This study analyzed the relationship between selecting critical errors (choices that would be dangerous to patients) and conventional test scores on a medical school certifying examination that included three item formats: regular and weighted multiple-choice questions and patient management problems. Data from a Clinical Certifying Examination administered to 279 seniors as the University of Illinois College of Medicine were analyzed. It was found that while there were significant negative correlations between test scores and number of critical errors made across the three different item formats, there were nonetheless students who passed the examination although they made a relatively large number of critical errors. Implications for teaching and testing are discussed. Four tables present data on critical errors. (SLD)