A test written the night before follows what is quick to ask: the first chapters, the definitions, the topic its writer likes best. A test blueprint settles the balance before the first question exists.
It is a small table. Topics run down the side, levels of thinking across the top, and every cell holds a share of the points. Many texts call it a table of specifications.
EdProcto, which publishes this guide, puts "Points" and "Topic (for the results by topic)" on each marked question in its builder, so the table can be counted before the sitting and read back after it. This guide is for a classroom or course test; a certification exam has its own. The sources were read on 10 October 2026: most come from medical schools, and those read only as abstracts are marked.
What is a test blueprint, and what is it for?
A test blueprint is the plan of a test, made before its questions: what will be asked, on which topic, at what level of thinking and for how many points. Its purpose is content validity, the claim that the test samples what was taught, in the proportions and at the level it was taught.
Coderre et al. (2009; abstract read) describe the aim as agreement between three things: the learning objectives, the learning experiences and the content of the evaluation. For Bridge et al. (2003; abstract read) the blueprint links the subject matter delivered in teaching to the questions on the test. Fives and DiDonato-Barnes (2013; abstract read) make the classroom case: the table frames a teacher's decisions while building a test and strengthens the judgements that rest on it.
The rows stop a test leaving out part of the course, the first of the two threats in our guide to validity. The columns stop a course taught as problem solving from being tested as definitions.
What a blueprint does not settle
It does not make a question good, and it does not prove that a question sits at the level its cell names: teachers often disagree about levels, as questions above recall reports. Nor is there one correct design. The 22 studies reviewed by Abdellatif et al. (2024) shared the same stages (purpose, domains, outcomes, format, weights) and no standard table.
What goes wrong in a test written without a blueprint?
The test drifts towards whatever is easiest to ask.
- Level. Marso and Pigge (1988; abstract read) collected 175 tests written by school teachers and rated the questions. Most worked mainly at the knowledge level; the mathematics tests were the exception.
- Planning. AlFallay (2018; abstract read) surveyed 199 teachers of English in Saudi schools. By their own report they usually prepared tests with no plan made beforehand, and the language points that lend themselves to testing came first.
- Coverage. Gill and Sen (2020; abstract read) set five years of one university's microbiology papers against the weights a department's staff judged right. Many topics were over-represented or under-represented across the five years. Of the questions, 8%, 20% and 8% sat at the first three levels of the revised taxonomy, and 64% were judged ambiguous.
Of the curriculum administrators at 144 medical schools surveyed by Bridge et al. (2003), most called blueprinting very important and few required it.
How do you weight the topics?
Start from teaching time, adjust for importance, and write both figures down. We start with time because it is what the students were given.
Abdellatif (2023) weighted each content area by the time spent teaching it, its relative value, how often it is used in practice and how much later learning depends on it. Abdellatif and Al-Shahrani (2019) used the number of learning objectives in each theme together with its contact hours.
The exact recipe may matter less than having one. Abdellatif and Al-Shahrani (2019) built two examinations for one course by two blueprint methods, of 21 and 30 questions, each sat by 40 students. Average difficulty (0.6 and 0.57), discrimination (0.38 and 0.36) and reliability (0.72 and 0.75) did not differ significantly. The groups were small, and the paper labels them inconsistently, so we do not match figures to methods.
A working rule for importance
Multiply a topic's hours by 1 as the default, and by up to 1.5 where later units depend on it. That is our rule, not a published formula. Keep it modest, so that judgement adjusts the teaching time without replacing it.
Weight in points, not in questions
A weight is a share of the points. Ten one-point questions and two five-point questions weigh the same, so a count of questions misleads as soon as questions differ in value.
How do you weight the levels of thinking?
Take the shares from the outcomes: count how many ask students to recall, to apply and to evaluate, and give each level that share of the points. Abdellatif and Al-Shahrani (2019) split their three levels this way, by the number of outcomes at each.
Use three levels. Those authors call the six levels of Bloom's taxonomy relatively difficult to apply in building an examination, and most studies in the Abdellatif et al. (2024) review sorted questions into three.
A cell may be empty: where no outcome asks students to evaluate a topic, write a zero.
How do you turn the weights into points and questions?
Multiply each topic's share by the test's points, split the total by the levels' shares, fill the cells so that rows and columns both add up, then divide each cell by the points one question of its kind carries.
The example is an illustration with invented figures: a 50-point test on a unit called Markets, taught in 25 hours of an introductory economics course. The teacher gives elasticity a multiplier of 1.5, because next term's units rest on it.
| Topic | Hours | Importance | Product | Share | Of 50 | Points |
|---|---|---|---|---|---|---|
| Demand, supply and equilibrium | 10 | 1 | 10 | 35.7% | 17.86 | 18 |
| Elasticity | 6 | 1.5 | 9 | 32.1% | 16.07 | 16 |
| Market failure | 5 | 1 | 5 | 17.9% | 8.93 | 9 |
| Government intervention | 4 | 1 | 4 | 14.3% | 7.14 | 7 |
| Total | 25 | 28 | 100% | 50.00 | 50 |
Take the second row: its product, 9, is 32.1% of 28, and 32.1% of 50 points is 16.07. By hours alone the four rows would have had 20, 12, 10 and 8 points.
The unit lists 20 outcomes: 8 ask for recall, 8 for application and 4 for evaluation, which is 40%, 40% and 20%, or 20, 20 and 10 points. A recall question is worth 1 point, an application question 2 and an evaluation question 5. Each cell gives its points and, in brackets, its questions: the points divided by the points one question carries.
| Topic | Recall | Apply | Evaluate | Points | Questions |
|---|---|---|---|---|---|
| Demand, supply and equilibrium | 8 (8) | 10 (5) | 0 | 18 | 13 |
| Elasticity | 5 (5) | 6 (3) | 5 (1) | 16 | 9 |
| Market failure | 2 (2) | 2 (1) | 5 (1) | 9 | 4 |
| Government intervention | 5 (5) | 2 (1) | 0 | 7 | 6 |
| Total | 20 (20) | 20 (10) | 10 (2) | 50 | 32 |
The last two columns show why points are the unit. Government intervention has six questions and market failure four, yet market failure carries 9 points to 7.
Check the minutes
Fix the length by the time, then test the plan against it. Allowing 45 seconds for a recall question, 2 minutes for an application question and 6 for an evaluation question, the test needs 15 + 20 + 12 = 47 minutes. Those allowances are ours: time your own questions. How many questions a quiz needs gives the least a topic needs before its score means much.
What did studies find about tests written with and without a blueprint?
Modest support, mostly from surveys, and almost no experiments. Abdellatif et al. (2024) searched two databases, found 487 records and kept 22 studies, nearly all from the health professions. Eight relied mainly on surveys, many written by the instructors themselves and with no clear evidence of validity. Six compared test statistics with and without a blueprint. The reviewers' own summary: "a lack of experimental studies was observed".
Two of the comparisons:
- Abdellatif (2023) compared two years of one course's examination, 25 multiple-choice questions each: 72 students sat the paper written without a blueprint and 71 the paper written with one. The blueprinted paper covered a wider range of difficulty, and no two of its questions depended on each other, as one pair did on the earlier paper. Mean scores were close (20.1 and 19.4 of 25), and the reliability of the students' scores was no higher with the blueprint (0.49 against 0.58). The abstract calls the differences in scores, person separation and item reliability non-significant. Only students who passed were analysed, the cohorts differed, and each paper's questions were drawn from a bank.
- Ahmad and Hamed (2014; abstract read) compared two groups in an imaging module, one examined with a blueprint and one without. The abstract reports statistically higher scores, satisfaction and achievement of the outcomes for one group, with better test statistics, and concludes in the blueprint's favour. It gives no figures.
The one randomised experiment we know of is reported in the certification guide.
What we could not find
We found no study that gave school classes a blueprinted test and an unplanned one and compared what the scores said. The case for a blueprint remains an argument about sampling, with supportive evidence around it.
How do you check a finished test against the blueprint?
Count the points in every cell of the finished test and set them beside the plan. A blueprint on file is no guarantee that the test follows it: Eweda et al. (2020; abstract read) built a tool to measure the match, and by the account in Abdellatif et al. (2024) it reports the share of questions that experts judge to fit.
- Label every question with its topic and level. In EdProcto, type the topic in "Topic (for the results by topic)", worded the same way each time, and the value in "Points".
- Total each row and column, and subtract the plan.
- Ask a colleague to label the questions blind. Where the two of you disagree, read the question again.
- Repair by swapping, one question out and one in, until every difference is zero or has its reason written beside it.
- Read the scores by topic afterwards. In EdProcto's results, "By topic" gives each topic's score; it is part of the full analysis, on the Educator plan.
The first draft of the Markets test, counted (invented figures again):
| Row or column | Planned | Draft | Difference |
|---|---|---|---|
| Demand, supply and equilibrium | 18 | 23 | +5 |
| Elasticity | 16 | 13 | −3 |
| Market failure | 9 | 7 | −2 |
| Government intervention | 7 | 7 | 0 |
| Recall | 20 | 27 | +7 |
| Apply | 20 | 18 | −2 |
| Evaluate | 10 | 5 | −5 |
The draft drifted the usual way, towards the first topic and towards recall, and the evaluation question on market failure was never written. The repair takes out five recall questions on demand and supply and three on market failure, and puts in that evaluation question, with one recall and one application question on elasticity: eight points out, eight in.
How does EdProcto help with a blueprinted test?
EdProcto is the best tool for a blueprinted class test because the two labels a blueprint needs, points and topic, sit on every question, and the scores come back by topic weighted by points. The example shows why that matters: counted by questions, two of its topics change places.
Building. In the builder each marked question has "Points", in steps of half a point, "Difficulty" ("Easy", "Medium" or "Hard") and "Topic (for the results by topic)". Choose each cell's shape from the templates: multiple choice for recall or application, a short answer or choose and justify for evaluation.
Drafting by cell. Procto AI's pane takes one topic's notes under "Or paste the material", with "How many" (1 to 40) and a slider, "Difficulty · 5 / 10". Its caption reads "Recall and recognition" from 1 to 3, "Application and analysis" from 4 to 7 and "Evaluation and synthesis" from 8 to 10: three bands for three columns. Ask for a cell's count and a few spares, and treat the level as a request: read each draft and decide its cell yourself.
Reading back. "By topic" lists every topic you named with its number of questions and the share of its points the class earned, so a five-point question counts five times a one-point one, as the blueprint intends. From five sittings, a sentence beginning "Reteach before moving on:" names the topics under 60%.
Plans. Free gives ten quizzes a month for thirty students a quiz, and Procto AI writes 20 questions a month. Tutor takes 100 students, opens all 111 templates and raises that to 50. Educator takes 500 students and adds the full analysis, "By topic" included. See pricing.
Quick questions
Is a table of specifications the same as a test blueprint?
Yes. Both names mean the table that sets topics against levels of thinking, with a weight in every cell.
How many questions should each topic get on a test?
Its share of the points, divided by the points one question carries. In the example, elasticity's 16 points became nine questions.
Should students see the blueprint?
Usually: the topics, levels and weights, never the questions. Students given blueprints in an inorganic chemistry course used them to organise revision around the heavily weighted objectives and took a positive view of the exams; exam scores did not rise (Young et al., 2019; abstract read).
Where does EdProcto show the score for each topic?
In the results, under "By topic", part of the full analysis on the Educator plan (pricing). Each topic's score is weighted by points.
References
- Abdellatif, H. (2023). Test results with and without blueprinting: Psychometric analysis using the Rasch model. Educación Médica, 24(3), Article 100802. https://www.elsevier.es/en-revista-educacion-medica-71-articulo-test-results-with-without-blueprinting-S1575181323000128
- Abdellatif, H., Alsemeh, A. E., Khamis, T., & Boulassel, M.-R. (2024). Exam blueprinting as a tool to overcome principal validity threats: A scoping review. Educación Médica, 25(3), Article 100906. https://www.elsevier.es/en-revista-educacion-medica-71-articulo-exam-blueprinting-as-tool-overcome-S1575181324000214
- Abdellatif, H., & Al-Shahrani, A. M. (2019). Effect of blueprinting methods on test difficulty, discrimination, and reliability indices: Cross-sectional study in an integrated learning program. Advances in Medical Education and Practice, 10, 23–30. https://doi.org/10.2147/AMEP.S190827
- Ahmad, R. G., & Hamed, O. A. (2014). Impact of adopting a newly developed blueprinting method and relating it to item analysis on students' performance. Medical Teacher, 36(Suppl. 1), S55–S61. https://pubmed.ncbi.nlm.nih.gov/24617786/
- AlFallay, I. S. (2018). Test specifications and blueprints: Reality and expectations. International Journal of Instruction, 11(1), 195–210. https://eric.ed.gov/?id=EJ1165235
- Bridge, P. D., Musial, J., Frank, R., Roe, T., & Sawilowsky, S. (2003). Measurement practices: Methods for developing content-valid student examinations. Medical Teacher, 25(4), 414–421. https://pubmed.ncbi.nlm.nih.gov/12893554/
- Coderre, S., Woloschuk, W., & McLaughlin, K. (2009). Twelve tips for blueprinting. Medical Teacher, 31(4), 322–324. https://pubmed.ncbi.nlm.nih.gov/18937095/
- Eweda, G., Bukhary, Z. A., & Hamed, O. (2020). Quality assurance of test blueprinting. Journal of Professional Nursing, 36(3), 166–170. https://pubmed.ncbi.nlm.nih.gov/32527639/
- Fives, H., & DiDonato-Barnes, N. (2013). Classroom test construction: The power of a table of specifications. Practical Assessment, Research, and Evaluation, 18, Article 3. https://digitalcommons.montclair.edu/educ-fdns-facpubs/39/
- Gill, J. S., & Sen, S. (2020). Blueprinting of summative theory assessment of undergraduate medical students in microbiology. Medical Journal Armed Forces India, 76(2), 207–212. https://pubmed.ncbi.nlm.nih.gov/32476720/
- Marso, R. N., & Pigge, F. L. (1988). An analysis of teacher-made tests: Testing practices, cognitive demands, and item construction errors (ED298174). ERIC. https://eric.ed.gov/?id=ED298174
- Young, K. J., Lashley, S., & Murray, S. (2019). Influence of exam blueprint distribution on student perceptions and performance in an inorganic chemistry course. Journal of Chemical Education, 96(10), 2141–2148. https://eric.ed.gov/?id=EJ1231315