Natural Environment Teaching: A Practitioner's Guide to NET, Incidental Teaching, and Blended NET-DTT Programming

What NET actually is — and what it isn't

For most BCBAs, "NET" is shorthand for a teaching arrangement with three properties: (1) the establishing operation is captured from the environment or contrived in advance rather than instructed by an adult; (2) the response form is mand-, tact-, intraverbal-, or play-shaped within an ongoing routine rather than at a discrete trial table; and (3) the consequence is a reinforcer that is functionally tied to the response — the bubble jar opens because the child manded "open," the next car is delivered because the child manded "more cars" (Frampton et al., 2024). Frampton and colleagues' clinical tutorial on capturing and contriving establishing operations is the cleanest contemporary statement of this architecture: practitioners actively engineer momentary EOs (preferred items in transparent closed containers within reach, interrupted chains during routine sequences, varied access intervals across the day) and respond with incidental-teaching steps when the learner emits the target (Frampton et al., 2024). The procedure is not less rigorous than DTT — it is differently structured. Discrimination work, prompting hierarchy, error-correction logic, and mastery criterion all still apply; they are simply embedded in a routine the learner already wants to be inside.

NET sits inside the NDBI family

LaMarca and LaMarca's ADDIE-based programming guide places incidental teaching, pivotal response training, enhanced milieu teaching, and reciprocal imitation training under the umbrella of Naturalistic Developmental Behavioral Interventions (NDBIs), all of which share child-led activity, natural reinforcement contingent on target responses, and embedded teaching trials inside ongoing social routines (LaMarca et al., 2024). The practical implication is that "NET" in current practice is not one procedure but a continuum: incidental teaching anchors the spontaneous-EO end, PRT and milieu approaches sit in the middle, and blended programming uses NET to generalize what DTT initially establishes (LaMarca et al., 2024). LaMarca and LaMarca are explicit that NDBI-variant selection should be matched deliberately to the learner — preferences, mastery rate, prerequisite repertoire — and not assumed (LaMarca et al., 2024). They also flag an operational risk that's easy to miss in naturalistic formats: because the reinforcer is intrinsic to the activity, a child who plays with the same train track for 30 minutes can quietly drift below the EO, so reinforcer rotation needs to be programmed even within "natural" play (LaMarca et al., 2024).

The mand-training literature is where NET is densest

McCammon, Wolfe, and Check's PRISMA-guided scoping review of 54 mand-training studies (1,489 children, 1980–2022) is the single best summary of how NET gets operationalized across the published mand corpus (McCammon et al., 2024). Roughly half of the reviewed studies embed mand instruction in NET-style arrangements, most commonly defined as child-led activity periods, loose stimulus control, and response prompting in everyday play or leisure contexts, with the classroom or therapy room serving as the typical physical setting (McCammon et al., 2024). The review's uncomfortable headline is that only about half of the studies supplied enough procedural detail to code all 25 environmental variables the authors targeted — meaning replicability across instructors and settings is genuinely limited even within the published literature (McCammon et al., 2024). The implication for practice is that lesson plans and data sheets need to operationalize the ecological variables the published reports often gloss: room arrangement, activity type, agent of instruction, prompting style, EO source (captured vs. contrived), and mastery criterion. Without that, NET drifts toward whatever the technician happened to do today, and a second clinician cannot replicate the procedure that made progress.

Generalization and incidental learning — the central theoretical claim

The traditional argument for NET is that skills taught inside the routines and reinforcers that will maintain them should generalize and maintain better than skills established at a table. The corpus is partly supportive and partly cautionary about that claim. Welsh and colleagues taught a perspective-taking component skill — tacting what another person currently sees, hears, tastes, feels, or smells — to three children with autism in family homes and community settings using NET with multiple-exemplar instruction and error correction; the children acquired the skill, and it generalized to untrained sensory modalities, untrained stimuli, and untrained people without additional formal probing Welsh et al. (2019). Anderson and Wiskow's telehealth instructive-feedback study with a 9-year-old showed that secondary, unsolicited targets were acquired alongside primary tact and intraverbal targets within the naturalistic teaching exchange — "bonus" learning that DTT alone typically does not produce without explicit programming (Anderson & Wiskow, 2025). Olaff and Holth pushed this further: in an alternating-treatments comparison with four typically developing preschoolers, mixed-operant instruction (listener + speaker responses interleaved on the same stimuli) produced rapid bidirectional naming, while incidental probing alone produced only listener responses (Olaff & Holth, 2025). Read together, these papers support the generalization claim — but with a procedural caveat: NET that consists only of adult probes during play under-delivers compared with NET that explicitly programs multiple operants on the same stimulus in context (Olaff & Holth, 2025). The "naturalness" of the arrangement is necessary but not sufficient.

Be evidence-honest about the strength of the generalization claim

Most of the NET evidence base is single-subject experimental design with small samples — typically 2–6 learners, often in the same clinic or classroom — and the field does not have a head-to-head NET-vs-DTT comparative effectiveness trial of the kind that would settle the generalization question prospectively Welsh et al. (2019) (Anderson & Wiskow, 2025). The McCammon scoping review documents the same limitation at the corpus level: half of the mand-training studies are NET-flavored, but the way "natural environment" is operationally defined varies enough that quantitative comparison across studies is difficult (McCammon et al., 2024). The honest read is that NET reliably produces acquisition for the targets it has been used to teach, that generalization within the NET arrangement is well-documented case by case, and that the strong form of the claim — "skills taught with NET generalize and maintain better than the same skills taught with DTT" — is supported by the logic of the procedure and individual SCED demonstrations, not by group-level comparative trials Welsh et al. (2019) (Olaff & Holth, 2025).

Language and communication outcomes

Most NET outcome data, in both research and practice, sit inside language and communication targets. Frampton and colleagues operationalize three NET-compatible mand procedures specifically — incidental teaching, the interrupted-chain procedure (ICP), and mand-training during pre-session pairing — and argue that mand frequency is most reliably increased when EOs are captured or contrived rather than relying on adult-instructed trials (Frampton et al., 2024). Welsh's perspective-taking work targets a verbal-behavior repertoire (tacting others' sensory experiences) that is hard to evoke at a table because the relevant sensory event is the natural one happening in the room Welsh et al. (2019). Anderson and Wiskow's telehealth tact and intraverbal demonstration shows that instructive feedback embedded in NET produces secondary-target acquisition without extra trials (Anderson & Wiskow, 2025). Hanney, Carr, and LeBlanc's auditory-tact study introduces a useful cautionary finding for NET-style language work: when sounds were paired with visual cues during tact instruction, acquisition accelerated, but the pairing disrupted previously mastered object-name tacts unless trials for both skills were interspersed Hanney et al. (2019). The implication is that "naturalistic" presentation does not automatically protect prior repertoires; interspersal and explicit maintenance probes still belong inside NET programming Hanney et al. (2019). Chang and colleagues' acquisition-criterion comparison — individual-item versus set-of-10 mastery — was conducted in tabletop conditions rather than NET, but the finding that per-item mastery accelerated tact acquisition argues that NET data systems can borrow this rule: track each target individually and advance when the target itself meets criterion, rather than waiting for a set to fill (Chang et al., 2024).

Data collection in NET — the practitioner's hardest problem

The single most flagged practitioner challenge in NET is data: how do you collect interpretable data inside a 20-minute play sequence without breaking the play? Raulston and colleagues' methodology paper makes the cleanest argument the field has produced on this question (Raulston et al., 2024). They argue that traditional trial-count metrics — borrowed from DTT — are inadequate for capturing child play and engagement during NET, because the granularity of a "trial" is the wrong unit for a fluid play routine, and the act of bracketing each behavior into a trial undermines the naturalistic structure that makes NET work (Raulston et al., 2024). Their recommended replacements are practical: simple frequency counts of targeted play actions (e.g., the number of pretend food items "cut" during a kitchen-play routine), 10- or 15-second momentary time-sampling for engagement states, and a 1–5 engagement rating scale that an RBT can apply at the end of an interval without freezing the activity (Raulston et al., 2024). The behavior-management rule they imply is procedural: data collection has to support implementation, not constrain it. If a measurement system requires the technician to interrupt the routine, the routine is no longer NET — it is a covert DTT arrangement with worse instructional control.

The corpus also gives practical templates for what to count. McCammon's scoping review documents that mand-training NET studies most commonly track mand frequency, mand topography (vocal, sign, AAC), and prompt level over the session (McCammon et al., 2024). Tryggestad and colleagues' Norwegian preschool staff-training study reinforces that a brief checklist (the PMT, in their study) lets teachers self-monitor the number and quality of contrived mand opportunities each day, sustaining naturalistic teaching without extra materials Tryggestad et al. (2025). The synthesis is direct: replace trial counts with frequency or interval data for child engagement, keep per-target plus/minus tallies for discrete language operants, and add a brief end-of-session checklist for whether the planned EOs were captured or contrived during the session (Raulston et al., 2024). Chang and colleagues' tact-acquisition findings argue for individual-item mastery tracking even inside NET; an RBT can record a per-target plus/minus tally on a clipboard during play without breaking the activity (Chang et al., 2024). ### Staff training matters more in NET than in DTT

NET's procedural integrity is harder to maintain than DTT's because the structure is loose and the teaching opportunities are emergent (Overstreet et al., 2025). Overstreet, Harvey, and May's multiple-baseline study used the teach-back method — having behavior therapists explain NET steps back to the trainer after written instructions, with brief vocal/corrective feedback only as needed — and brought all autism-clinic therapists to 100% NET procedural integrity across two consecutive sessions (Overstreet et al., 2025). The headline is the efficiency: a 5-minute teach-back step plus targeted feedback was sufficient, suggesting full BST packages are not always required for NET (Overstreet et al., 2025). Tryggestad and colleagues took the same logic to mainstream Norwegian preschools: 17 weeks of in-service coaching plus brief BST produced large, immediate gains in teachers' ability to contrive evocative situations, capture motivational operations, and otherwise enhance NET opportunities — but benefits began fading within 10 weeks of coaching withdrawal Tryggestad et al. (2025). The maintenance signal is the operationally important finding: NET fidelity decays without ongoing coaching, and supervision plans need to budget periodic boosters, not assume a single training cycle holds Tryggestad et al. (2025).

The physical environment is itself a teaching variable

Jackson and colleagues' multiple-baseline study at a day-services agency for adolescents and adults with ASD and other disabilities (n = 15) is the cleanest demonstration that the physical layout of the room is a contingency operating on the learner Jackson et al. (2022). Re-arranging furniture and materials — clearer sight lines, clustered seating, defined activity zones, storage moved to room perimeters — reliably increased the proportion of time clients were actively engaged with tasks or staff, with engagement gains in the 25–40% range across rooms Jackson et al. (2022). The implication for NET is concrete: before adding more complex teaching procedures, run an environmental check. U-shaped seating, accessible-but-not-free reinforcers, and visible activity zones can pull engagement up before a single mand trial is run, and they reduce the prompting load on staff once instruction begins Jackson et al. (2022). Frampton and colleagues' EO-engineering recommendations operate at the same scale: transparent closed containers within reach, motivating items in defined locations, and routine sequences with deliberate interruption points are physical-environment design choices as much as teaching procedures (Frampton et al., 2024).

Generalization beyond verbal behavior — community and life-skills targets

NET extends meaningfully beyond mand and tact training. Torelli and colleagues' kindergarten life-skills evaluation embedded teaching trials wholly within play centers, the cafeteria, the playground, and PE rather than at isolated tables, contriving evocative situations during naturally favored activities to target requesting, gaze shifting, and name-response skills for three kindergarteners with developmental disabilities (Torelli et al., 2026). The skills were acquired, maintained, and generalized to untrained settings, and the authors' procedural notes are instructive: rotate stimulus items and settings to promote generalization without structured tabletop trials, allow natural consequences rather than contrived tokens, and use a brief natural buffer (≥3 minutes after group instruction) before inserting embedded trials so the trials retain authenticity (Torelli et al., 2026). Dixon and colleagues' immersive-VR pedestrian-safety study went the other direction — using a structured non-natural training environment (a head-mounted display) to teach a community-essential skill that then generalized to real-street probes for all six children with autism who tolerated the equipment, with one participant withdrawn for HMD intolerance Dixon et al. (2020). The implication for NET programming is to favor direct community probes whenever safety allows, reserving structured environments for prerequisite or risk-sensitive teaching where in-vivo errors would be costly.

Family and home implementation

NET maps naturally onto family routines, and the corpus contains explicit guidance for caregiver-implemented variants. Szabo and colleagues' COVID-era practical paper describes how parents can weave academic, language, and social goals into ongoing family activities — turning hikes or chores into "Math Mountain," using shared reading or emotion-based guessing games during walks, treating activity completion itself as the contingent reinforcer rather than reaching for tokens Szabo et al. (2020). The paper is conceptual rather than experimental, but it reads as a useful caregiver-coaching template precisely because it operationalizes "embed in routines" at a level a parent can actually implement Szabo et al. (2020). Frampton's tutorial on capturing and contriving EOs translates directly to home implementation — kitchens, bathrooms, and bedrooms are full of natural EO opportunities that a coached parent can convert into mand trials within minutes (Frampton et al., 2024).

Telehealth NET

Telehealth NET is feasible but under-studied. Lindgren and colleagues' multiple-baseline comparison (n = 7 young children with ASD) of telehealth versus in-person DTT incorporated NET as part of the overall package — separate targets were taught with NET in both modalities — though the study did not report a NET-specific modality comparison (Lindgren et al., 2024). The procedural takeaway is that NET targets can be paired with DTT delivered remotely without compromising the naturalistic component, provided the caregiver presents reinforcers and the NET activities use items already available in the home (Lindgren et al., 2024). Anderson and Wiskow's instructive-feedback demonstration was conducted via telehealth with a single 9-year-old learner and showed that secondary targets were acquired alongside primary tact and intraverbal targets, indicating the incidental-learning benefits typical of in-person NET extend to telehealth delivery — at least at the case-study level (Anderson & Wiskow, 2025). Both studies argue for caregiver-implemented NET as a default mode in telehealth, with the BCBA coaching from a remote camera rather than running trials directly (Lindgren et al., 2024) (Anderson & Wiskow, 2025).

The NET literature lives mostly at the single-subject-experimental-design layer with one scoping review and a band of conceptual and methodology work around the edges (McCammon et al., 2024) Welsh et al. (2019). There are no group-level comparative-effectiveness trials of NET versus DTT in the available corpus (McCammon et al., 2024).

Systematic and scoping reviews. McCammon, Wolfe, and Check's PRISMA-guided scoping review of 54 mand-training studies (1980–2022) is the most comprehensive synthesis available, cataloguing 25 environmental variables across the corpus and documenting that NET-style arrangements appear in roughly half of mand-training studies, with substantial variability in operational definition (McCammon et al., 2024).

Single-subject experimental designs. Most NET evidence is SCED Welsh et al. (2019). Welsh and colleagues' multiple-baseline (n = 3) demonstrates NET-based perspective-taking acquisition with cross-modality, cross-stimulus, and cross-person generalization in family and community settings Welsh et al. (2019). Anderson and Wiskow's adapted alternating-treatments design (n = 1) shows secondary-target acquisition during telehealth tact and intraverbal instruction with instructive feedback (Anderson & Wiskow, 2025). Olaff and Holth's alternating-treatments design (n = 4 typically developing preschoolers) isolates mixed-operant instruction as the active ingredient for bidirectional naming, with incidental probing alone producing only listener responses (Olaff & Holth, 2025). Torelli and colleagues' kindergarten multiple-baseline (n = 3) demonstrates NET-based life-skills acquisition with maintenance and generalization to untrained school settings (Torelli et al., 2026). Jackson and colleagues' multiple-baseline across settings (n = 15) demonstrates that environmental rearrangement alone increases engagement 25–40% in an agency context Jackson et al. (2022). Hanney, Carr, and LeBlanc's multiple-baseline across sounds (n = 2) is methodologically smaller but operationally important: it documents that compound auditory-visual presentations during tact instruction can disrupt previously mastered object-name tacts unless interspersed Hanney et al. (2019). Chang and colleagues' alternating-treatments design (n = 3) in a tabletop context is included for the acquisition-criterion finding that translates to NET data systems (Chang et al., 2024). Dixon and colleagues' multiple-baseline (n = 6, with one withdrawal for HMD intolerance) demonstrates VR-to-natural-environment generalization for pedestrian safety Dixon et al. (2020). Lindgren and colleagues' multiple-baseline (n = 7) compares telehealth and in-person DTT with NET embedded in both arms (Lindgren et al., 2024).

Staff-training and implementation studies. Overstreet, Harvey, and May's multiple-baseline study of teach-back for NET fidelity reached 100% procedural integrity across autism-clinic therapists (specific N not extractable from available chunks) (Overstreet et al., 2025). Tryggestad and colleagues' delayed multiple-probe across three Norwegian mainstream preschools is the largest NET-fidelity training study in the corpus, with a 17-week coaching cycle and a 10-week follow-up that documented decay without continued support Tryggestad et al. (2025).

Methodology and conceptual papers. Raulston, Ousley, Hinton, and Ramirez's measurement paper is the field's most cited contemporary statement on data collection during NET, arguing against trial-count metrics and for frequency, interval, and rating-scale alternatives (Raulston et al., 2024). Frampton and colleagues' clinical tutorial operationalizes how to capture and contrive establishing operations within NET — the cleanest current reference for incidental teaching, the interrupted-chain procedure, and pre-session pairing (Frampton et al., 2024). LaMarca and LaMarca's ADDIE-based programming guide places NET within the broader NDBI family and argues for deliberate format selection by learner (LaMarca et al., 2024). Szabo and colleagues' family-routines paper translates NET principles into caregiver-implementable activities Szabo et al. (2020).

Bottom line. The convergent picture across these tiers supports the operational claims this page makes — that NET reliably produces acquisition for mand, tact, intraverbal, perspective-taking, and life-skills targets; that environmental design is itself a teaching variable; that staff training is shorter than commonly assumed but maintenance requires ongoing coaching; and that data systems should match the procedure rather than be borrowed from DTT (Frampton et al., 2024) Jackson et al. (2022) (Overstreet et al., 2025) (Raulston et al., 2024). The evidence is thinner for the strong generalization claim that NET produces durably better learner outcomes than DTT in head-to-head trials, where the existing literature is single-subject demonstrations and procedural variability documented in one scoping review Welsh et al. (2019) (McCammon et al., 2024).

The NET decisions a senior practitioner makes are not "NET or DTT" so much as "which format, for which target, in which setting, with which data system." A defensible logic, drawn directly from the corpus:

1. New mand or tact target, learner has prerequisite imitation and approach repertoires. Default to NET with captured or contrived EOs. Position preferred items in transparent closed containers within reach; insert deliberate interruptions in routine chains; reinforce immediately with the natural consequence (Frampton et al., 2024).
2. Target requires precise stimulus discrimination, dense trial count, or systematic prompt fading on novel arbitrary stimuli. Use DTT to establish the discrimination, then move to NET to generalize and maintain. Per-item mastery tracking accelerates acquisition and converts cleanly to NET maintenance probes (Chang et al., 2024).
3. Bidirectional naming or other "incidental learning" target. Mixed-operant instruction (listener + speaker responses interleaved on the same stimuli) outperforms incidental probing alone. Build the multi-operant structure explicitly even inside NET — do not rely on emergent practice (Olaff & Holth, 2025).
4. Perspective-taking or other socially situated repertoire. Run NET in family and community settings with multiple exemplars across modalities, stimuli, and people; the generalization properties of the SCED literature justify investing in cross-context exemplars from the start Welsh et al. (2019).
5. Life skills or community participation target. Embed teaching trials inside the actual setting — playground, cafeteria, community sidewalk — rather than in a clinic analog (Torelli et al., 2026). Rotate stimuli and settings to promote generalization without explicit probes, and insert a 3-minute natural buffer after group instruction to protect authenticity (Torelli et al., 2026).
6. Engagement is low and instruction is loading staff prompting beyond capacity. Run an environmental scan first. Re-arrange seating, accessibility of materials, and activity zones before adding teaching procedures. Engagement gains of 25–40% are achievable from layout changes alone Jackson et al. (2022).
7. Auditory or compound-stimulus tact targets. Pair sounds with visual cues for faster acquisition, but intersperse trials with already-mastered visual tacts to prevent interference with previously acquired repertoires Hanney et al. (2019).
8. Caregiver-implemented programming at home. Use the family's own routines as the EO source — kitchens, bathrooms, walks, shared reading. Treat activity completion itself as the contingent reinforcer; reach for tokens only when natural contingencies do not maintain responding Szabo et al. (2020) (Frampton et al., 2024).
9. Telehealth delivery. Pair NET with caregiver-presented reinforcers using items already in the home; the BCBA coaches from the remote camera while the caregiver runs the trials, and incidental-learning benefits typically extend to telehealth at the case level (Lindgren et al., 2024) (Anderson & Wiskow, 2025).
10. Data system for NET sessions. Replace trial counts with frequency or interval data for child engagement, keep per-target plus/minus tallies for discrete language operants, and add a brief end-of-session checklist for whether the planned EOs were captured or contrived. Do not collect data in any way that requires breaking the routine (Raulston et al., 2024).
11. Staff training for NET. Start with written instructions plus a 5-minute teach-back; layer brief vocal/corrective feedback only if the technician does not reach mastery. Schedule ongoing coaching boosters — fidelity decays within 10 weeks of withdrawal (Overstreet et al., 2025) Tryggestad et al. (2025).
12. Risk-sensitive community skill (pedestrian safety, etc.). Use a structured non-natural environment (VR or simulated setting) to establish prerequisite responding before community probes, but plan the in-vivo generalization probe explicitly so the simulation is not treated as the terminal goal.

Clinic-based NET

Clinic-based NET is where most published procedures were developed and validated. Frampton and colleagues' EO-engineering tutorial reads as a clinic-implementation manual: arrange the therapy room with transparent closed containers within reach, prepare interrupted-chain materials for routine sequences, and track which EO sources were captured versus contrived during the session (Frampton et al., 2024). Anderson and Wiskow's clinic-based instructive-feedback work demonstrates that NET formats embed cleanly with secondary-target programming (Anderson & Wiskow, 2025). Olaff and Holth's preschool-clinic alternating-treatments design supports the operational rule that mixed-operant blocks should be programmed explicitly inside NET rather than left to emerge from probing (Olaff & Holth, 2025). Overstreet, Harvey, and May's autism-clinic teach-back study gives the staff-training template — a 5-minute teach-back step plus brief vocal feedback brings entry-level therapists to 100% NET procedural integrity, which is faster and cheaper than a full BST package and operationally sustainable in clinic supervision rotations (Overstreet et al., 2025).

School NET

School NET works inside ongoing classroom routines rather than at instructional tables. Tryggestad and colleagues' Norwegian preschool study is the field's strongest current demonstration that mainstream-classroom teachers can be coached to contrive evocative situations and capture motivational operations using a low-dose in-service package, with large gains in NET implementation following 17 weeks of coaching plus brief BST Tryggestad et al. (2025). The maintenance signal is the operationally important part: implementation gains began fading within 10 weeks of coaching withdrawal, meaning school-based NET requires sustained supervisory infrastructure rather than a one-time training Tryggestad et al. (2025). Torelli and colleagues' kindergarten life-skills study demonstrates that NET can target requesting, gaze shift, and name-response skills inside play centers, the cafeteria, the playground, and PE rather than at isolated tables, and that those skills generalize and maintain across settings (Torelli et al., 2026). Jackson and colleagues' day-services finding extends to school: U-shaped seating, accessible materials, and defined activity zones drive engagement up before any teaching procedure is layered on, and the same logic applies in K–12 classrooms Jackson et al. (2022).

Home and family-implemented NET

Home NET maps onto the family's existing routines, with the BCBA coaching the caregiver rather than running the trials directly Szabo et al. (2020). Szabo and colleagues' practical paper operationalizes the approach with concrete templates — "Math Mountain" hikes, gratitude walls, emotion-based guessing games during walks, treating activity completion itself as the contingent reinforcer rather than reaching for external tokens Szabo et al. (2020). Frampton's EO-engineering tutorial translates directly to home implementation: kitchens (closed snack cabinets, transparent jars, interrupted snack-prep chains), bathrooms (manded water-on, manded soap), and bedrooms (interrupted dressing chains, manded help with shoes) are dense EO environments that a coached parent can convert to mand trials within minutes (Frampton et al., 2024). Welsh's perspective-taking work was conducted in family homes and community settings and is direct evidence that NET-driven generalization is achievable when training happens inside the family's actual social contexts, not in a clinic analog Welsh et al. (2019).

Telehealth NET

Telehealth NET is feasible with caregiver-mediated implementation (Lindgren et al., 2024). Lindgren and colleagues showed that NET targets can be embedded alongside DTT in both telehealth and in-person service-delivery packages without compromising the naturalistic component, provided caregivers present reinforcers and use items already in the home (Lindgren et al., 2024). Anderson and Wiskow's case-level demonstration extends this to instructive-feedback variants of NET — secondary-target acquisition during telehealth tact and intraverbal sessions (Anderson & Wiskow, 2025). Both studies argue for caregiver-implemented NET as the default mode in telehealth, with the BCBA coaching from a remote camera and field-scale replications still needed (Lindgren et al., 2024) (Anderson & Wiskow, 2025).

Adult and residential NET

Jackson and colleagues' agency study with adolescents and adults with ASD and other disabilities (n = 15) is the corpus's strongest evidence that NET-style environmental design generalizes to adult-services settings Jackson et al. (2022). Re-arranging furniture and materials raised engagement 25–40% across rooms, and the authors argue that environment checks can replace or supplement 1:1 prompting, reducing staff fatigue while keeping clients in natural interaction loops Jackson et al. (2022). The implication for adult-services BCBAs is that NET in this population often starts with environmental redesign rather than direct teaching procedures — and that the leverage from layout changes is large enough to justify running the multiple-baseline design within your own program before adding more complex teaching components Jackson et al. (2022).

The most common practitioner question about NET is when to use it instead of (or alongside) discrete trial training, and the corpus supports a structural rather than ideological answer (Frampton et al., 2024).

Use DTT when: the target requires high trial density to establish (e.g., novel arbitrary discriminations, precise prompt fading on stimuli that don't exist in the natural environment), the learner doesn't yet have the prerequisite imitation, attending, or approach repertoire to engage in NET productively, or the target is a precision skill where errors during acquisition would be costly to correct later (Chang et al., 2024). Per-item mastery criteria accelerate acquisition relative to set-based criteria, and that's easier to enforce in DTT than in NET (Chang et al., 2024).

Use NET when: the target is mand, tact, intraverbal, social, perspective-taking, or life-skills work where the natural environment already supplies the EO and the natural reinforcer; generalization to untrained people, settings, or stimuli is part of the goal rather than a separate phase; or the learner's engagement and motivation rise materially when instruction happens inside their preferred activity (Frampton et al., 2024) Welsh et al. (2019) (Torelli et al., 2026).

Use blended NET-DTT when: DTT establishes the discrimination and NET generalizes it — the dominant pattern in modern early-intensive programming (LaMarca et al., 2024). The Lindgren and colleagues telehealth study illustrates the structure operationally — separate targets taught in DTT and NET inside the same overall package, with telehealth and in-person delivery both viable (Lindgren et al., 2024). LaMarca and LaMarca's NDBI framing supports the same pattern at the program level: format selection should be matched to the target and the learner, not chosen as a global commitment (LaMarca et al., 2024). Chang and colleagues' acquisition-criterion finding bridges the two formats — per-item mastery tracking is operationally compatible with NET and accelerates acquisition relative to set-based mastery (Chang et al., 2024).

The decision is target-driven, not philosophical: a clinician running only DTT loses the generalization advantage NET supplies, and a clinician running only NET loses the trial density and discrimination control DTT supplies (Chang et al., 2024) Welsh et al. (2019).

• Drift to "play with no plan." This is the most common failure mode. NET is not free play with a behavior analyst nearby. It requires planned EOs, a defined target list, an end-of-session checklist for whether EOs were captured or contrived, and per-target data. If a session looks like adult-led play without those elements, it is not delivering the procedure (Frampton et al., 2024) (Raulston et al., 2024).
• Reinforcer mismatch. "Natural" does not mean "any consequence available in the room." The reinforcer has to be functionally tied to the target response — opening the bubbles is the reinforcer for the open-bubbles mand, not generic praise (Frampton et al., 2024). Plan reinforcer rotation explicitly even within preferred play; satiation creeps in faster than practitioners expect (LaMarca et al., 2024).
• Missed teaching opportunities. Captured EOs are by definition emergent, and emergent opportunities get missed when staff are watching the data sheet instead of the learner. The Tryggestad coaching study suggests that brief self-monitoring checklists for "contrived mand opportunities per session" pull staff attention back to the EO surface during the session Tryggestad et al. (2025).
• Trial-count data inside NET. Borrowing DTT data systems into NET creates a procedural contradiction: you cannot run a fluid play routine while bracketing each behavior into a discrete trial, so replace trial counts with frequency, interval, and per-target tally systems that survive the routine (Raulston et al., 2024).
• Probing only, with no mixed-operant practice. Olaff and Holth's finding is direct: incidental probing alone produces only listener responses for novel stimuli, while mixed-operant instruction produces full bidirectional naming. NET that consists of adult questions during play under-delivers on incidental learning (Olaff & Holth, 2025).
• Ignoring interference between previously mastered and new operants. Hanney and colleagues' compound-stimulus auditory-tact finding documents that "naturalistic" presentation does not protect prior repertoires automatically. Intersperse new compound-stimulus trials with already-mastered visual tacts to prevent regression Hanney et al. (2019).
• One-shot staff training. Tryggestad and colleagues showed that NET fidelity gains began fading within 10 weeks of coaching withdrawal, so plan periodic boosters into the supervision schedule rather than assuming a single training cycle will hold Tryggestad et al. (2025).
• Skipping the environmental scan. Jackson and colleagues' agency study showed engagement gains of 25–40% from layout changes alone — failing to run an environmental check leaves easy engagement gains on the table and inflates the prompting load on staff Jackson et al. (2022).
• Using NET for targets DTT would establish faster. Precision discriminations on novel arbitrary stimuli, dense trial-count tact acquisition under tight prompt fading, and prerequisite imitation-deficit cases often progress faster with DTT first, then NET for generalization (Chang et al., 2024). Choosing NET globally for ideological reasons costs the learner acquisition speed.
• Treating "naturalistic" as synonymous with "unstructured." The Frampton tutorial is explicit on this: capturing and contriving EOs is a structured procedure that requires preparation, materials, and integrity checks. The looseness is in the trial structure, not in the planning (Frampton et al., 2024).

• Severe problem behavior that interrupts NET sessions repeatedly. Refer for a function-based assessment and behavior plan before continuing skill-acquisition work; NET cannot operate when severe behavior dominates the session ecology.
• Suspected medical or sensory substrate driving low engagement. Behavior that may involve pain, sleep, sensory, or seizure issues should be referred for medical evaluation before assuming environmental redesign or NET procedures will produce engagement gains.
• Learner who consistently fails to engage with any contrived EO across multiple staff and settings. When the basic capture/contrive moves don't produce mand attempts after a defined trial period, refer for preference assessment, prerequisite-skill review, and possibly a more structured DTT acquisition phase before returning to NET (Frampton et al., 2024).
• Staff cannot reach mastery on NET fidelity after two coaching cycles. When teach-back plus brief feedback does not produce ≥80% procedural integrity, refer the case to a more experienced clinician or restructure training rather than running underpowered NET sessions (Overstreet et al., 2025) Tryggestad et al. (2025).
• Risk-sensitive community skill where direct in-vivo training is unsafe. Pedestrian safety, water safety, and similar skills with consequential errors should be established in structured or simulated environments first, then probed in the natural environment under controlled conditions.

The honest read of the corpus is that the operational, practitioner-facing claims on this page sit on solid SCED and tutorial evidence, while the comparative-effectiveness layer is essentially absent Welsh et al. (2019) (McCammon et al., 2024). The field does not yet have a head-to-head NET-vs-DTT randomized or quasi-experimental trial that would settle the strong generalization claim, and that gap is the largest single opportunity for the next five years of research (McCammon et al., 2024).

Several smaller-scale gaps are more tractable. McCammon and colleagues' scoping review documents that only about half of mand-training studies report enough procedural detail to code 25 environmental variables, meaning even within the published literature the NET arrangement is variably defined — and a consensus reporting standard for NET would meaningfully improve cross-study comparison and replication (McCammon et al., 2024). The Raulston and colleagues measurement paper makes a strong argument against trial-count data in NET, but it is conceptual rather than empirical, and a prospective comparison of trial-count, frequency, interval, and rating-scale data systems against the same NET targets would convert their recommendation into evidence-grade procedure (Raulston et al., 2024). Tryggestad and colleagues' fidelity-decay finding deserves a longer-horizon replication: 10 weeks is short, and the actual half-life of NET implementation gains across multiple agency contexts is unknown Tryggestad et al. (2025). Telehealth NET has been demonstrated only at the case-study and small-multiple-baseline level; a multi-site replication of caregiver-implemented telehealth NET, with explicit modality comparison, would tell the field whether the in-person evidence base actually transfers (Lindgren et al., 2024) (Anderson & Wiskow, 2025). Jackson and colleagues' environmental-manipulation finding deserves replication outside agency settings — particularly in K–12 and home contexts where layout choices are constrained differently — and the active ingredients of the rearrangement (sight lines, clustering, accessibility) have not been isolated Jackson et al. (2022).

1. Define NET by structure, not setting. The procedure is captured/contrived EO + natural-routine response form + functionally tied reinforcer — not "anywhere outside the table" (Frampton et al., 2024). A session at a play center that lacks planned EOs and per-target data is not NET; a session at a clinic table that captures the learner's reach for a closed jar and reinforces with the contents is NET.
2. Engineer the environment before you engineer the procedure. Transparent closed containers within reach, U-shaped seating, defined activity zones, and routine sequences with deliberate interruption points are the physical-environment moves that make NET productive (Frampton et al., 2024) Jackson et al. (2022). Layout changes alone produce 25–40% engagement gains in agency and day-service contexts.
3. Choose format by target, not by ideology. Use DTT for precision discriminations on arbitrary stimuli; use NET for mand, tact, intraverbal, perspective-taking, and life-skills work where natural reinforcers are abundant; use blended NET-DTT when DTT establishes the discrimination and NET generalizes it (LaMarca et al., 2024) (Chang et al., 2024).
4. Replace trial counts with frequency, interval, and rating-scale data. Trial-count data systems borrowed from DTT break the routine that makes NET work. Use simple tallies of targeted play actions, 10–15-second momentary time-sampling for engagement, and a 1–5 engagement rating scale at the end of intervals (Raulston et al., 2024).
5. Track per-target acquisition individually. Per-item mastery criteria accelerate acquisition relative to set-of-10 mastery in tabletop work, and the same logic applies to NET — keep a per-target plus/minus tally and advance when each target meets criterion (Chang et al., 2024).
6. Program mixed operants explicitly, not just probes. Incidental probing alone produces listener responses; mixed-operant instruction produces full bidirectional naming. Build listener + speaker practice on the same stimulus into NET — don't rely on emergent practice (Olaff & Holth, 2025).
7. Capture and contrive EOs deliberately. Position preferred items in transparent closed containers within reach; insert routine interruptions during chains the learner already does; vary access intervals across the day to keep MOs active (Frampton et al., 2024).
8. Default to caregiver-implemented NET in home and telehealth. Coach the parent through the family's existing routines — kitchens, walks, shared reading, dressing chains — and treat activity completion itself as the contingent reinforcer when natural contingencies maintain responding Szabo et al. (2020) (Frampton et al., 2024) (Lindgren et al., 2024).
9. Train staff with teach-back first, full BST only if needed. A 5-minute teach-back plus brief vocal/corrective feedback brings entry-level therapists to 100% NET fidelity — faster and cheaper than a full BST package (Overstreet et al., 2025).
10. Budget NET fidelity boosters into supervision. Implementation gains decay within 10 weeks of coaching withdrawal. Schedule periodic 30-minute coaching cycles in the classroom or clinic; do not assume one training holds Tryggestad et al. (2025).
11. Intersperse new compound-stimulus trials with already-mastered tacts. "Naturalistic" presentation does not automatically protect prior repertoires — compound auditory + visual tact training can disrupt mastered visual tacts unless interspersed Hanney et al. (2019).
12. For perspective-taking and other socially situated targets, run NET across people, settings, and stimulus modalities from the start. The generalization properties of the SCED literature justify investing in cross-context exemplars during acquisition rather than treating generalization as a separate phase Welsh et al. (2019).
13. For life-skills targets, embed teaching trials inside the actual setting. Playground, cafeteria, sidewalk — the site is the SD, and inserting a ≥3-minute natural buffer after group instruction preserves authenticity of the embedded trials (Torelli et al., 2026).
14. Plan reinforcer rotation explicitly even within preferred play. Satiation creeps in faster than practitioners expect, especially in NET where the reinforcer is intrinsic to a single activity. Rotate reinforcers within and across sessions (LaMarca et al., 2024).
15. Use a brief end-of-session NET checklist. Did the planned EOs get captured or contrived? Did mixed-operant practice happen for each language target? Did per-target data get collected without breaking the routine? Three yes/no items pull staff attention back to procedural integrity without requiring a full BST recheck Tryggestad et al. (2025) (Raulston et al., 2024).

Primary research synthesized in this guide. DOIs link to the original source.

• Raulston, T. J., Ousley, C. L., Hinton, E. M., & Ramirez, A. M. (2024). Beyond trial counts: Considerations for measuring play and engagement during early intervention for autistic children. Behavior Analysis in Practice, 17, 1216–1227. https://doi.org/10.1007/s40617-024-01002-3 https://doi.org/10.1007/s40617-024-01002-3
• Overstreet, D., Harvey, C., & May, M. E. (2025). Using the teach-back method to improve staff implementation of naturalistic environmental teaching. Journal of Organizational Behavior Management. https://doi.org/10.1080/01608061.2025.2594400 https://doi.org/10.1080/01608061.2025.2594400
• Frampton, S. E., Davis, C. R., Meleshkevich, O., & Axe, J. B. (2024). A clinical tutorial on methods to capture and contrive establishing operations to teach mands. Behavior Analysis in Practice, 17, 1270–1282. https://doi.org/10.1007/s40617-024-00985-3 https://doi.org/10.1007/s40617-024-00985-3
• Welsh, F., Najdowski, A. C., Strauss, D., Gallegos, L., & Fullen, J. A. (2019). Teaching a perspective‐taking component skill to children with autism in the natural environment. Journal of Applied Behavior Analysis, 52(2), 439-450. https://doi.org/10.1002/jaba.523 https://doi.org/10.1002/jaba.523
• LaMarca, V. J., & LaMarca, J. M. (2024). Using the ADDIE model of instructional design to create programming for comprehensive ABA treatment. Behavior Analysis in Practice, 17, 371–388. https://doi.org/10.1007/s40617-024-00908-2 https://doi.org/10.1007/s40617-024-00908-2
• Jackson, J. W., Dunkel-Jackson, S. M., Dixon, M. R., & Szekely, S. (2022). Effectiveness of Environmental Manipulation to Enhance Engagement and Ecological Validity at an Agency Serving Individuals With Autism Spectrum Disorder and Other Disabilities. Behavior Analysis in Practice, 15(4), 1390-1395. https://doi.org/10.1007/s40617-019-00392-z https://doi.org/10.1007/s40617-019-00392-z
• McCammon, M. N., Wolfe, K., & Check, A. R. (2024). A review of the environmental variables included in mand training interventions. The Analysis of Verbal Behavior, 40, 345–378. https://doi.org/10.1007/s40616-024-00211-9 https://doi.org/10.1007/s40616-024-00211-9
• Anderson, B. K., & Wiskow, K. M. (2025). Acquisition of secondary targets during tact and intraverbal instruction with instructive feedback. The Analysis of Verbal Behavior, 41, 40–56. https://doi.org/10.1007/s40616-025-00215-z https://doi.org/10.1007/s40616-025-00215-z
• Tryggestad, H., Eldevik, S., Eikeseth, S., Strømgren, B., & Kingsdorf, S. (2025). Staff Training to Improve the Preschool Educational Environment for Children with Special Needs. Behavior Analysis in Practice. https://doi.org/10.1007/s40617-025-01123-3 https://doi.org/10.1007/s40617-025-01123-3
• Torelli, J. N., Snyder, S. K., Griffin, M. L., Hellemn, M. P., Cagliani, R. R., & Morgan, G. A. (2026). Life skills evaluation in a kindergarten special education classroom. Journal of Applied Behavior Analysis, 59(1), e70041. https://doi.org/10.1002/jaba.70041 https://doi.org/10.1002/jaba.70041
• Lindgren, N. A., Higbee, T. S., Osos, J. A., Nichols, B., & Campbell, V. E. (2024). Comparing the effectiveness of discrete trial training delivered via telehealth and in‑person on skill acquisition. Behavior Analysis in Practice, 17, 783–795. https://doi.org/10.1007/s40617-023-00855-4 https://doi.org/10.1007/s40617-023-00855-4
• Olaff, H. S., & Holth, P. (2025). Acquisition of incidental bidirectional naming: Isolating the effects of probing and mixed-operant instruction. The Analysis of Verbal Behavior, 41, 200–234. https://doi.org/10.1007/s40616-025-00221-1 https://doi.org/10.1007/s40616-025-00221-1
• Szabo, T. G., Richling, S., Embry, D. D., Biglan, A., & Wilson, K. G. (2020). From Helpless to Hero: Promoting Values-Based Behavior and Positive Family Interaction in the Midst of COVID-19. Behavior Analysis in Practice, 13(3), 568-576. https://doi.org/10.1007/s40617-020-00431-0 https://doi.org/10.1007/s40617-020-00431-0
• Chang, H., Nainani, T., & Kim, J. Y. (2024). Comparison of acquisition criteria applied to individual and sets of tacts: A systematic replication. Behavior Analysis in Practice, 17, 632–637. https://doi.org/10.1007/s40617-024-00933-1 https://doi.org/10.1007/s40617-024-00933-1
• Dixon, D. R., Miyake, C. J., Nohelty, K., Novack, M. N., & Granpeesheh, D. (2020). Evaluation of an Immersive Virtual Reality Safety Training Used to Teach Pedestrian Skills to Children With Autism Spectrum Disorder. Behavior Analysis in Practice, 13(3), 631-640. https://doi.org/10.1007/s40617-019-00401-1 https://doi.org/10.1007/s40617-019-00401-1
• Hanney, N. M., Carr, J. E., & LeBlanc, L. A. (2019). Teaching children with autism spectrum disorder to tact auditory stimuli. Journal of Applied Behavior Analysis, 52(3), 733-738. https://doi.org/10.1002/jaba.605 https://doi.org/10.1002/jaba.605