DTT vs NET: Two Teaching Formats, One Set of Principles

Discrete trial training and natural environment teaching are both applied behavior analysis. They use the same principles — reinforcement, prompting, fading, motivating operations — arranged differently. Treating them as rival philosophies is a category error that costs learners progress.

Discrete trial training

DTT breaks a skill into small units taught in repeated, structured trials. Each trial has a clear four-part structure:

1. Discriminative stimulus — the instruction or cue.
2. Response — what the learner does, possibly prompted.
3. Consequence — reinforcement for correct responding, or a correction procedure.
4. Intertrial interval — a brief pause that separates one trial from the next.

Trials are typically massed, adult-directed, and run with materials chosen by the instructor.

What DTT is good at

• Building new discriminations. High trial density means many learning opportunities per minute, which matters enormously in early acquisition.
• Skills with no natural motivating operation. Receptive identification of community signs does not come up naturally often enough to teach incidentally.
• Clean data. Discrete trials produce discrete data. Trial-by-trial accuracy and prompt level are easy to record and easy to interpret.
• Consistency across staff. A well-written DTT program can be run the same way by different people, which matters on a real caseload with turnover.

Where DTT struggles

• Generalization. A response taught at a table with one instructor and one set of cards frequently fails to occur anywhere else. This is DTT's well-documented weakness and it must be programmed around, not assumed away.
• Spontaneity. DTT teaches responding to instructions. It does not naturally produce initiation.
• Motivation. Reinforcers in DTT are often arbitrary relative to the response, which makes the behavior dependent on those reinforcers being present.

Natural environment teaching

NET embeds teaching in ongoing activities and routines, following the learner's motivation and using consequences naturally related to the response. A learner reaching for bubbles is taught to request bubbles, and the reinforcer is bubbles.

What NET is good at

• Generalization and maintenance. Because teaching happens across settings, people and materials with natural reinforcers, responses are far more likely to occur outside teaching.
• Motivation. The learner already wants the item, so the motivating operation is in place before the trial starts rather than having to be contrived.
• Spontaneous initiation. NET teaches mands in conditions where mands actually occur.
• Social validity. It looks like play and interaction, which matters to families and to the learner's dignity.

Where NET struggles

• Trial density. Far fewer learning opportunities per minute. For a learner who needs hundreds of trials to acquire a discrimination, waiting for natural opportunities is slow.
• Coverage. Some targets simply do not arise naturally at useful frequency.
• Data. Harder to capture cleanly, and consequently more often captured poorly or after the fact.
• Staff skill. NET is harder to run well. It requires the instructor to recognize and capitalise on opportunities in real time, which is a higher skill ceiling than delivering a scripted trial.

How to combine them

The practical model most programs converge on is acquire in DTT, generalize in NET — with the caveat that this is a default rather than a rule.

1. Acquisition in DTT. Use massed trials to establish the discrimination efficiently, with prompt fading and a clear mastery criterion.
2. Loosen within DTT. Before leaving the table, vary instructors, materials, instruction phrasing, and setting. This is where most programs stop too early.
3. Transfer to NET. Contrive opportunities in natural routines, then capture naturally occurring ones. Reinforce with the natural consequence.
4. Probe for generalization in untrained settings and with untrained people. Generalization that has not been measured has not been demonstrated.

Some targets skip step one entirely. Manding is generally better taught in NET from the start, because a mand taught in a contrived context with an arbitrary reinforcer is not really a mand.

The data problem

The reason many programs drift toward DTT-heavy delivery is not clinical. It is that DTT produces data easily and NET does not.

NET data is genuinely harder to capture: opportunities arise unpredictably, the instructor's hands are usually occupied, and recording after the fact turns data into recollection. The common failure is that NET happens but goes unrecorded, so the graph shows only table work and the clinical record understates what the learner can actually do.

What helps is a capture method fast enough to use mid-activity — single-tap recording against the same targets as the DTT program, on a device the instructor already has in hand, so a naturally occurring opportunity gets scored in the moment rather than reconstructed at the end of the session. When NET data costs the same as DTT data, the balance of teaching stops being driven by measurement convenience.

What the evidence supports

Both formats have substantial empirical support, and comparative research generally finds each has the advantages described above — DTT stronger on acquisition rate, NET stronger on generalization and spontaneous use. The reasonable reading is that format should be chosen per target and per learner rather than adopted as a house style.

A program that is 100% DTT is likely under-serving generalization. A program that is 100% NET is likely slow on acquisition for learners who need high trial density. Neither is a defensible default.

The summary

DTT gives trial density, clean data and staff consistency, and it is weak on generalization and initiation. NET gives generalization, motivation and spontaneity, and it is weak on trial density, coverage and data capture. Use DTT to acquire, loosen within DTT before you leave it, transfer to NET, and probe generalization deliberately. Teach mands in NET from the start. And make sure NET data is as easy to record as DTT data, or measurement convenience will quietly decide your clinical model.