UWorld vs AMBOSS: What the Research Says About QBank Selection

“Should I use UWorld or AMBOSS?” is one of the most common questions in Step 1 forums, group chats, and advising sessions. Students agonize over this choice as if the QBank itself determines their score. The published data tells a different story.

What matters is not which QBank you pick. What matters is how many questions you complete, how you review them, and when you start. The research on this point is consistent across multiple studies and institutions.

The Question Volume Data

Burk-Rafel et al. (2017) surveyed 274 medical students at the University of Michigan about their Step 1 study behaviors. The results were published in Academic Medicine (PubMed 29065026). Students completed an average of 3,597 practice questions during their study period. They studied an average of 11 hours per day over 35.3 days.

The critical finding: after controlling for MCAT scores, preclinical exam performance, and self-identified score goals, the number of practice questions completed remained a significant predictor of Step 1 scores(adjusted R² = 0.56, p < .001). Attempting more questions was associated with higher scores regardless of the student's baseline ability.

Notice what that finding does not say. It does not say “UWorld questions predicted scores but AMBOSS questions did not.” It does not distinguish between question sources at all. The variable that mattered was total questions completed.

Guilbault et al. (2020) confirmed the pattern in Medical Science Educator (PMC8368851). Practice questions completed and preclinical grades were positively correlated with Step 1 scores. Study hours per day? The correlation was r = 0.07. Effectively zero. The total time spent sitting at a desk predicted almost nothing about the final score. The number of questions worked through did.

What UWorld and AMBOSS Actually Offer

Before looking at more data, it helps to understand what each platform does differently. Both are question banks. Both cover Step 1 material. The differences are in structure and supplementary features.

• ▸UWorld has roughly 2,700+ Step 1 questions. Explanations are long and detailed, often several paragraphs per answer choice. The platform is question-first: you do blocks, review explanations, repeat. No integrated reference library.
• ▸AMBOSS has a comparable question count plus an integrated knowledge library with highlighted terms that link to reference articles. Questions tend to be shorter. The platform functions as both a QBank and a searchable reference.

Ahmad and Perkins (2023) analyzed the dermatology content across both UWorld and AMBOSS Step 1 question banks in Clinical and Experimental Dermatology (PubMed 36763765). They found 655 dermatology questions across both platforms, with differences in topic distribution and image inclusion rates. The study was content-focused, not outcomes-focused, but it illustrates an important point: the QBanks are not identical. They emphasize different topics and present material differently. Neither one covers everything perfectly.

The QBank Access Effect

Schwartz et al. (2017) published a quasi-experimental study in Teaching and Learning in Medicine (PubMed 29240455) that is relevant here. At the University of Illinois, a student-initiated program provided second-year students with early access to a question bank (specifically UWorld) partway through the academic year, along with peer-led review sessions.

After the program launched, average Step 1 scores increased by 8.82 points compared to preintervention cohorts (p < .001). The first-time pass rate rose by 8%. When controlling for age, sex, MCAT, and undergraduate GPA, program participants scored 6.57 points higher and had 3.08 times the odds of passing Step 1.

The intervention was not “UWorld vs. AMBOSS.” The intervention was early, structured access to a question bank plus peer-led review. The platform was incidental. The behavior change was what moved scores.

Spaced Repetition and QBank Synergy

Gilbert et al. (2023) published a cohort study in Medical Science Educator examining Anki usage among 130 first-year medical students (PMC10403443). Of these, 78 used Anki for at least one exam and 52 did not. The result: Anki users scored 12.9% higher on the CBSE (Comprehensive Basic Science Exam), even after controlling for MCAT scores (p = 0.003).

Across all four exams in the study, Anki users scored significantly higher. The effect sizes ranged from 6.2% to 12.9%, with the largest gap on the CBSE, which most closely approximates USMLE Step 1 content.

This matters for the QBank question because Anki and QBanks serve complementary functions. QBanks test application of knowledge and expose gaps. Anki consolidates what you learn from those QBank reviews into long-term memory through spaced repetition. Neither alone is as effective as both together.

The practical implication: whichever QBank you choose, pair it with an active spaced repetition system. When you miss a question or learn a new concept from a QBank explanation, that information should feed into your Anki deck. The QBank identifies gaps. Anki fills them permanently. This approach is discussed further in our study schedule breakdown.

The Step 2 CK Data Extends the Pattern

Drake et al. (2021) studied Step 2 CK preparation strategies at Michigan State and published their findings in PRiMER (PMC8437322). Two strategies correlated with Step 2 CK scores in preliminary analyses: completing more practice questions and the proportion of a question bank completed. In hierarchical regression controlling for demographics and Step 1 performance, only completing more questions remained a significant predictor.

This replicates the Step 1 finding in a different exam context. For Step 2 CK, the same principle held: question volume predicted scores. The specific resource did not.

So Which One Should You Pick?

The honest answer: the research does not support a clear winner. No published head-to-head study has randomized students to UWorld vs. AMBOSS and compared Step 1 outcomes. That study does not exist. Every “UWorld is better” or “AMBOSS is better” claim you read online is anecdote, not evidence.

What the research does support is a set of behaviors that predict higher scores, regardless of QBank choice:

1. ▸Complete a high volume of questions.Burk-Rafel's data showed 3,597 as the average. Higher-scoring students did more. Guilbault confirmed that question count, not study hours, correlated with scores.
2. ▸Start before dedicated.77% of students in the Burk-Rafel study had already begun their QBank during the preclinical year. Schwartz's intervention confirmed that early QBank access raises scores by nearly 9 points.
3. ▸Review thoroughly, not quickly. The Burk-Rafel study measured question completion, not question speed. The benefit comes from working through explanations, understanding wrong answers, and building connections. Rushing through 100 questions with surface-level review is not what the high-scoring students did.
4. ▸Pair QBank work with spaced repetition.Gilbert's data showed a 12.9% CBSE score advantage for Anki users. The QBank surfaces what you do not know. Anki makes sure you do not forget what you learned from the review.
5. ▸Track your errors by category. Not every wrong answer is a content gap. Some are test-taking errors that require a different fix. The QBank is only useful if you are diagnosing your mistakes correctly.

A Practical Framework for Choosing

Since the evidence does not declare a winner, the decision comes down to personal workflow and budget. Here are the real tradeoffs:

• ▸If you learn best from long, detailed explanations and want maximum depth per question, UWorld's explanation style suits that.
• ▸If you want an integrated reference library you can search during study (not just during QBank review), AMBOSS provides that.
• ▸If budget allows, many students use both. Start AMBOSS during the preclinical year (it pairs well with curriculum study) and add UWorld for dedicated. This maximizes total question exposure, which is the variable the research says actually matters.
• ▸If you can only afford one, pick whichever you will actually finish. An incomplete QBank is a worse outcome than the “wrong” QBank completed cover to cover.

The Bottom Line

Students spend hours debating UWorld vs. AMBOSS when the published data points to a simpler conclusion: the QBank brand matters far less than the study behaviors surrounding it. Complete a high volume of questions. Start before dedicated. Review carefully. Layer in spaced repetition. Track your error patterns.

The students who struggle are rarely using the “wrong” QBank. They are using the right QBank with the wrong approach: doing questions passively, skimming explanations, studying hours without tracking what those hours produce. The evidence on study scheduling confirms this. Hours do not predict scores. Active question-based practice does.

Pick a QBank. Commit to finishing it. Build a system around it. That is what the research says works.

References

• Burk-Rafel J, Santen SA, Purkiss J. “Study Behaviors and USMLE Step 1 Performance: Implications of a Student Self-Directed Parallel Curriculum.” Academic Medicine, 2017; 92(11S):S67-S74. PubMed 29065026.
• Guilbault RWR, Lee SW, Lian B, Choi J. “Predictors of USMLE Step 1 Outcomes: Charting Successful Study Habits.” Medical Science Educator, 2020; 30(1):103-106. PMC8368851.
• Gilbert S, et al. “A Cohort Study Assessing the Impact of Anki as a Spaced Repetition Tool on Academic Performance in Medical School.” Medical Science Educator, 2023. PMC10403443.
• Schwartz LF, Lineberry M, Park YS, Kamin CS, Hyderi AA. “Development and Evaluation of a Student-Initiated Test Preparation Program for the USMLE Step 1 Examination.” Teaching and Learning in Medicine, 2017; 30(2):193-201. PubMed 29240455.
• Drake E, Phillips JP, Kovar-Gough I. “Exploring Preparation for the USMLE Step 2 Exams to Inform Best Practices.” PRiMER, 2021; 5:26. PMC8437322.
• Ahmad M, Perkins SH. “Learning dermatology in medical school: analysis of dermatology topics tested in popular question banks.” Clinical and Experimental Dermatology, 2023; 48(4):361-363. PubMed 36763765.