Three measures of person-fit (the extent to which an examinee's response pattern on a multiple-choice test is consistent with his ability as estimated by total score) were computed for students taking classroom tests under 12 different instructors at a comprehensive university. Supplementary questions on each test inquired concerning students' physical and mental states at the time of testing (e.g., health and extent of preparation), since it had been conjectured that these might be related to person-fit. Internal consistency reliability estimates were computed for the person-fit measures, and stability of these measures over time was evaluated for two sequences of tests over two groups of students. Only very weak evidence was found in support of the conjecture that poor person-fit might be associated with examinee circumstances at the time of testing, evidence far too weak to permit evaluating the measures with respect to each other. Other analyses suggested that poor person-fit typically arises from an examinee's unexpected responses to a very small proportion of items and that substantially different subsets of items account for this phenomenon from one examinee to another. Accordingly, internal consistency reliability estimates are probably inappropriate for person-fit measures. It was found that person-fit measures were highly unstable within examinees across several different tests in the same academic course. These results strongly question the potential usefulness of person-fit measures and call for studies to evaluate test-retest reliability of person-fit measures. (Author)