An item format incorporating pattern recognition was designed to assess medical students' abilities in the area of clinical diagnosis. A group of approximately 20 faculty members of five New England medical schools met in Worcester for half of a day to develop pattern recognition items. Teams of four to six physicians were assigned to work on particular topic areas that represent common chief complaints of patients. They developed a list of approximately 15 common diagnoses that relate to each of the topics. An item describing a patient by listing critical signs and symptoms was developed for each of the diagnoses. Approximately 300 items divided into 21 sets were developed and, subsequently, edited and reviewed by independent physicians before test administration. A modified Angoff procedure was used to set pass/fail standards for the set of items. A total of 336 fourth-year medical students from the five schools were tested using the items. The 21 sets of pattern recognition items were completed by between 112 and 332 examinees. Results indicate that: (1) students performed well on the items--the mean score was 82% correct, and almost 66% of the students passed at least 90% of the sets they took; (2) generalizability analyses indicated that performance in one topic area did not predict performance in other areas very well; and (3) 2 hours of testing time would be required to generate a reasonably reliable score. Four tables and eight figures are provided. (TJH)