Many schools seek to predict performance on national exams required for medical school graduation using prematriculation and medical school performance data. The need for targeted intervention strategies for at-risk students has led much of this interest. Assumptions that preadmission data and high stakes in-house medical exams correlate strongly with national standardized exam performance needs to be examined. Looking at prematriculation data for predicting USMLE Step 1 performance, we found that MCAT exam totals and math-science GPA had the best prediction from a set of prematriculation values (adjusted R 2 = 11.7 %) for step 1. The addition of scores from the first medical school exam improved our predictive capabilities with a linear model to 27.9 %. As we added data to the model, we increased our predictive values as expected. However, it was not until we added data from year 2 exams that we started to get step 1 prediction values that exceeded 50 %. Stepwise addition of more exams in year two resulted in much higher predictive values but also led to the exclusion of many early variables. Therefore, our best step 1 predictive value of around 76.7 % consisted of three variables from a total of 37. These data suggest that the preadmission information is a relatively poor predictor of licensure exam performance and that including class exam scores allows for much more accurate determination of students who ultimately proved to be at risk for performance on their licensure exams. The continuous use of this data, as it becomes available, for assisting at-risk students is discussed (251).