Residents have limited time and much to learn. Retrieval practice—studying using exam-style review questions—is a powerful educational tool for aggregate groups of learners, but it has not been tested for individual graduate medical education (GME) trainees.

We conducted a retrospective observational study examining how individual residents’ self-directed retrieval practice affected their learning. We used linear regression models to compare the number of exam-style multiple-choice practice questions each resident answered per year to their scores on annual In-Training Exams (ITE).

We found that the higher a resident’s ITE score was at the start of a year, the fewer practice questions they answered that year for both first- (p = 0.023) and second-year (p = 0.020) trainees. Then, the more questions a resident answered in a year, the more their ITE score increased over that year for both first- and second-year trainees (p = 0.026 and 0.025, respectively). Residents’ prior ITE scores also independently predicted their subsequent ITE scores (p = 0.024 in the first and 0.007 in the second year), and the effect of their baseline scores was larger than the effect of the number of practice questions they answered.

Individual residents’ prior exam scores significantly predict their future exam scores, and their independent retrieval practice using exam-style review questions also significantly predicts the annual improvement in their scores. These findings add to the growing evidence supporting retrieval practice in GME, and they can inform individualized educational coaching for GME trainees.