Stimulus Discrimination Training in ABA: Examples & How to Track Data

Put three cards in front of a learner and say “touch car.” He taps the card on the left. Next trial, “touch shoe,” and he taps the left card again. Then “touch cup.” Left again. About a third of those answers will be right by pure luck, and if all you track is percent correct, the data can look like progress.

That’s the problem stimulus discrimination training in ABA is meant to solve. You want a learner who responds to the right cue and ignores the wrong ones, and you want a data sheet that can tell real learning from a lucky habit. Below you’ll find the core concepts, clinical examples, teaching steps, and a practical way to track data so you can trust what it’s telling you.

Quick summary

• Stimulus discrimination means a response is reinforced in the presence of one stimulus (the SD) and not in the presence of another (the S-delta).
• Most early learner programs, like receptive identification, use conditional discrimination.
• Percent correct alone can mislead you. Track prompt level, field size, and error patterns too.
• Set the mastery criterion before you start, and know what chance level looks like for your field size.

What stimulus discrimination training in ABA means

Stimulus discrimination training teaches a learner to respond differently depending on what’s in front of them. A response gets reinforced when a certain stimulus is present and doesn’t when it isn’t. After enough repetitions, that stimulus starts to control the behavior.

Two terms do most of the work here. The discriminative stimulus (SD) signals that a response will be reinforced. The stimulus delta (S-delta) signals that it won’t. A phone ringing is an SD for answering it. A silent phone is an S-delta.

Situation	SD	Response	S-delta	What happens in S-delta
Receptive identification	Instruction “touch red,” red card in the field	Touches red card	Blue card in the field	No reinforcement, trial re-presented
Responding to name	Adult says the learner’s name	Looks toward the adult	Adult says a different name	Looking isn’t reinforced
Requesting	Adult is holding the snack	Says “chips, please”	Adult has nothing in hand	Request isn’t reinforced in training
Crossing the street	Walk signal	Crosses	Don’t-walk signal	Waits

One quick note before moving on. Discrimination and generalization pull in opposite directions. Generalization means responding the same way across similar stimuli, like recognizing a dog whether it’s a poodle or a lab. Discrimination means responding differently across stimuli that matter, like telling a dog from a cat. Good programming needs both, which is why generalization shows up later in the teaching steps.

Simple vs. conditional discrimination

A simple discrimination has one SD. You respond when it’s present and not when it’s absent. Answering the phone when it rings is the classic case.

A conditional discrimination is trickier, because the right response depends on a combination of stimuli. In receptive identification, the red card is correct only when the instruction is “red.” Say “blue” and the same red card is now wrong. The learner has to track the instruction and the array at once.

That’s why receptive ID looks easy on a program sheet and turns out to be hard in a session. Matching, receptive labels, and most “find the one that…” targets are all conditional discriminations.

Examples of stimulus discrimination training in ABA

These are the targets you’ll see most often in clinic programs.

Receptive identification is the obvious one. The instruction is “touch car,” the field holds a car and two other items, and only the car gets reinforced.

Identical matching works the same way with a sample. The learner gets a cup card and places it with the matching cup, ignoring the shoe.

Responding to name falls here too. The name called is the SD, and other names or sounds are S-deltas. For many young learners this is the first discrimination worth teaching.

Requesting by audience is a quieter example. A learner who asks for the iPad should ask the adult who’s holding it, not a person across the room with empty hands. The adult holding the item is the SD.

Safety skills count as well. Reading a walk signal versus a don’t-walk signal, or stopping at a stop sign, are discriminations with real consequences.

Social targets use the same logic. Greeting someone who just arrived is reinforced, while greeting the wall isn’t, which sounds silly until you meet a learner who greets everyone at the same volume all day.

How to teach it

The procedure changes by program, but the shape is pretty consistent.

1. Pick targets and a starting field. Begin with two items that look and sound very different. Car and cup is a better pair than cup and cap.
2. Give the instruction the same way every time, with a clear response to make (touch, point, hand over).
3. Prompt errorlessly at first, usually with a most-to-least prompt, so the learner reaches reinforcement quickly and doesn’t practice mistakes.
4. Reinforce correct responses. When the learner picks the wrong item, don’t reinforce it, then re-present the trial with a prompt.
5. Rotate the target’s position every trial. A common rule of thumb is no more than two in a row in the same spot.
6. Fade prompts as soon as the learner is correct without them. Prompts that stay too long become the cue the learner is actually responding to.
7. Expand the field to three items, then more, and change the distractors.
8. Add new materials, people, and settings so the skill doesn’t live in one room with one therapist.

Mixing in a few mastered targets between new ones helps too. It keeps reinforcement flowing and gives the learner an easy win after a hard trial.

A worked example (illustrative, not a real client)

Say Leo is four. In baseline, his RBT runs “touch car,” “touch shoe,” and “touch cup” in a field of three, ten trials. Leo gets 3 of 10 right, and 8 of his 10 selections are the card on the left. That’s a position habit, not knowledge.

The plan: drop to a field of two (car and cup), rotate positions every trial, use a gestural prompt at first, and fade it as soon as he responds without it. After a few sessions he’s at 9 of 10 independent in the field of two. Then the field goes to three, and then new pictures and real objects get added.

The written goal might read: Leo will touch the named item from a field of three, with positions rotated, at 90% independent across three consecutive sessions, with two staff members and in two settings.

Notice that a position pattern sat in the baseline data the whole time. Only a data sheet that records which card was chosen would show it.

How to track data in stimulus discrimination training

The bare minimum is correct or incorrect on each trial. You’ll want more than that, because a single percent hides a lot.

What to record	What it tells you
Independent correct (percent)	Overall progress, without prompts
Prompt level on each trial	Whether the learner is truly independent or leaning on the prompt
Which item was selected on errors	Position bias, or a favorite item the learner always picks
Field size and distractors	How hard the trial really was
Staff member and setting	Early signs of generalization (or lack of it)
Sessions to criterion	Pace, and when it’s time to change the teaching

Know what chance looks like

With a field of two, guessing gets you about 50%. With a field of three, about 33%. If a learner is sitting at 40% in a field of three, that’s barely above guessing, and calling it “emerging” overstates it. Write chance level next to the goal so everyone reading the graph knows what the floor is.

Set the mastery criterion up front

Pick the criterion before you start teaching, not after you see the data. Something like 80% to 90% independent across two or three consecutive sessions is common, but your clinic and funder rules decide. Whatever you pick, add the number of staff and settings, so mastery means the skill works beyond one therapist in one room.

What to do when the graph goes flat

If data hold flat for several sessions, change something. Maybe the distractors are too similar to the target, maybe the prompt never faded, or maybe the reinforcer lost its pull. And if errors cluster on one position or one item, that’s a clue about the cause. Trend lines make this easier to see than a column of numbers.

Common mistakes

Position bias is the quietest one. If target placement isn’t rotated and the data sheet doesn’t capture where the learner pointed, the pattern stays invisible.

Teams also expand the field too fast. Moving from two items to five because the learner hit criterion once usually just produces a new pile of errors.

Teaching with one example is another trap. A learner who only sees one red card may be responding to that card’s bent corner rather than to “red.” Using several examples, in different materials, helps the skill transfer.

Prompt dependence sneaks in when a prompt that should fade stays in place. The learner gets correct answers, the graph climbs, and nobody notices the answers disappear when the prompt does.

Changing the instruction wording from trial to trial can muddy things early on. Keep it consistent until the discrimination is stable, then vary it.

And finally, treating prompted and independent responses as the same data point. They aren’t. A 90% that includes heavy prompting isn’t a 90%.

Keeping the data in one place

Most of what makes discrimination data useful (prompt level, field size, error patterns, trend) is also what makes it tedious on paper. That’s usually where good programs slip: the information is collected, but nobody can see it clearly in time to adjust.

MeasurePM’s clinical data collection is built for this kind of workflow, with real-time analytics and mobile access, so your team can record trials during the session and see the graph update right away. Its customizable programming lets you set flexible measurement types for skill acquisition, success criteria, trend detection, and automatic progression of targets and prompts. That fits discrimination programs well, since the field size and prompt level change as the learner moves along.

The same records feed documentation. Progress reports that tie each goal to session data make authorization requests easier to write, and connected ABA billing software keeps claims lined up with the services you delivered.

If you’d like more clinical and operational reading, the MeasurePM resource center has other ABA guides.

Frequently asked questions

What is stimulus discrimination training in ABA?

It’s a teaching procedure where a response is reinforced when one stimulus is present and not reinforced when another is. Over time the learner responds to the right cue and ignores the wrong ones. Common targets include receptive identification, matching, responding to name, and reading safety signals.

What’s the difference between SD and S-delta?

An SD signals that a response will be reinforced. An S-delta signals that it won’t. In “touch red,” the red card is the SD and the blue card is the S-delta. The learner learns to respond to the first and hold off on the second.

How is stimulus discrimination different from stimulus generalization?

Discrimination means responding differently to different stimuli. Generalization means responding the same way across similar stimuli. A learner who calls both a poodle and a lab a “dog” is generalizing. One who doesn’t call a cat a “dog” is discriminating. Good programs teach both.

How do you know when a learner has mastered a discrimination?

When independent correct responding meets the mastery criterion you set at the start, usually across several consecutive sessions, with more than one staff member and setting. Also check that errors aren’t clustering on one position or item, and that results sit well above chance for the field size.

What is conditional discrimination?

It’s when the correct response depends on a combination of stimuli. In receptive identification, touching the red card is right only if the instruction was “red.” The same card is wrong when the instruction is “blue.” Matching and most receptive language targets work this way.

Before your next discrimination program

Good discrimination training comes down to a few habits: clear targets, prompts that fade, positions that rotate, and data that records more than right or wrong. Decide the criterion up front, keep an eye on chance level, and read the error patterns as closely as the percent correct.

If you’d like to see how that data looks in a single system, you can book a MeasurePM demo.