Updated 2026-11-12

RBT® question bank: 1,016 questions, how they are written and checked

1,016 original questions: 315 free without sign-up, 701 in the Study plan. Every question is written for one of the 43 tasks in the 2026 outline, no question appears twice, and every wrong option is explained. With the Study plan, an AI coach helps you understand any question you answered. The Study plan is $49 one time, fully refunded if you finish at least one of its mocks and don't pass (refund terms).

The numbers on this page are counted from the bank each time the site is built.

What is in the bank

Free, no sign-up

SetQuestions
20-question check20
85-question mock exam85
Six domain quizzes120
Six confusable-pairs sets90
Free total315

In the Study plan

SetQuestions
Practice by domain234
Look-alike pairs, such as DRA vs. DRO212
Three more mock exams255
Study plan total701

By domain

DomainQuestions
Data Collection and Graphing169
Behavior Assessment123
Behavior Acquisition248
Behavior Reduction179
Documentation and Reporting139
Ethics158
All six1,016
  • Every task is covered. The 2026 Test Content Outline has 43 tasks. Each has between 11 and 47 questions; the study guide lists the count for every task.
  • No question appears twice. Each question sits in one set only, so a mock exam never repeats a question you met in a quiz or in practice by domain.
  • Every wrong option is explained. Each question has an explanation of the answer and a reason for each of the three other options.
  • Most are scenarios. 862 of the 1,016 put you in a session and ask what is happening or what to do.

How our questions are written and checked

As of November 2026.

  1. Built from the 2026 outline. We turned the BACB's Test Content Outline (3rd edition: 43 tasks, six domains) into a library of 203 concepts. Each one has a definition in our own words, the terms people confuse it with, and scenario examples.
  2. Written from that library. We write and refine every question for one task with Claude Fable 5.2 (Anthropic) and GPT-6 Astra (OpenAI), working only from the concept library.
  3. Styled after the BACB's published sample questions. We studied the retired sample questions in the RBT Handbook for length and wording. No stem here is longer than 84 words, and the correct answer is the longest option in only 14% of questions. We do not copy or adapt those samples, and we use no real exam content.
  4. Answered blind, three times. Every question goes to three other AI models without the answer key: models from OpenAI and Anthropic for every question, and a Google model as one of the three for 732 of them. A question is published only if all three reach the keyed answer and none finds a second defensible option. Across the bank that is 3,048 blind answers.
  5. Reviewed line by line. On November 5, 2026, GPT-6 Astra reviewed every question then in the bank, 981 of them. Each finding was checked, confirmed problems were fixed before launch, and 14 questions were removed because a model found two defensible answers.
  6. The checking itself was tested. An independent agent wrote 21 deliberately flawed questions of seven kinds and mixed them in with real ones. The checks caught all 21.
  7. Verified again after every rewrite. When we revised the wording and length of the bank against the BACB's sample questions, each revised question was answered blind again.
  8. Human-verified. A person on our team verifies every question, its answer and its explanations before it goes live.

If you think a question is wrong, email support@btexampractice.com and we check it against the sources.

One question in full

This is the first question of the free Data Collection and Graphing quiz, with everything you see after you answer it.

During a 50-minute clinic session, an RBT counted 15 instances of self-injury. The client's graph has a y-axis labeled "Responses per hour."

What value should the RBT plot?

  • A 15 per hour
  • B 30 per hour
  • C 12.5 per hour
  • D 18 per hour

The answer is D.

Summarizing data · task A.6

Why D is right

Rate is count divided by time observed, in the unit the graph uses. A 50-minute session is 50/60 of an hour, so 15 divided by 0.833 hours is 18 per hour; equivalently, 0.3 per minute times 60 is 18 per hour. Matching the unit on the y-axis keeps every point on the graph comparable.

Why the others are wrong

  • A 15 is the raw count for the session; the axis asks for a rate per hour, and the session was shorter than an hour.
  • B 30 comes from dividing 15 by 50 and multiplying by 100, which treats the rate like a percentage.
  • C 12.5 comes from dividing 15 by 1.2 instead of by 50/60 of an hour; the conversion is inverted.

The AI coach answers from our own reviewed knowledge base

The AI coach comes with the Study plan. Without it, you can try it on 3 answers.

  1. It works on a question you just answered. It sees the question, the option you picked, the reviewed explanation, and why each other option is wrong.
  2. It draws on our own knowledge base. This is the concept library the questions are written from: 203 concepts that we compiled from the 2026 Test Content Outline and reviewed, so its answers follow the current outline.
  3. It adds nothing of its own. The coach is told not to bring in facts, numbers or rules that are not in that material.
  4. It explains your mistake, not just the answer. If your pick was wrong, it says why that option looks tempting, using the common confusions recorded for the concept.
  5. It stays on the question. It will not predict your score or tell you whether you will pass.

A general AI assistant answers from whatever it has read, which may include older material. The AI coach is limited to the reviewed explanation and concept cards for the question in front of you.

How this compares with other practice sites

We checked 10 widely used RBT practice sites on November 11, 2026.

  • 4 still use the pre-2026 outline, in whole or in part.
  • 8 are free to use (one of them also sells a paid app) and 2 are paid. Ours has 315 free questions; the Study plan is $49 one time, fully refunded if you finish at least one of its mocks and don't pass.

How to check any practice test yourself

  1. Look at the domain names. The 2026 outline has six: Data Collection and Graphing, Behavior Assessment, Behavior Acquisition, Behavior Reduction, Documentation and Reporting, and Ethics. If you see "Measurement", "Skill Acquisition" or "Professional Conduct and Scope of Practice", the questions follow the old Task List.
  2. Count the tasks. The 2026 outline has 43.
  3. Check the weights. The 75 scored questions split 13 / 8 / 19 / 14 / 10 / 11 across the six domains.
  4. Check the mock exam. It should have 85 questions, 75 scored and 10 unscored, in 90 minutes.

Found a mistake?

Email support@btexampractice.com with the question and what you think is wrong. We check it against the sources and fix the question if you are right.

Find the domains to study first.

20 questions, weighted like the exam. About 15 minutes.