An analysis of the evaluation instruments of clinical clerkships from 54 medical schools was made. Instruments were classified as to purpose, format, and skills measured. Thirty-nine schools used a modified Likert format; a few schools also had a check list of adjectives or short answer questions. Nearly all instruments had some space for general comments. The most frequently evaluated concepts and skills were "knowledge,""getting along well with others,""hard worker,""ability,""dignity,""history-taking," and "performance." Several principles of the design of evaluation instruments were discussed. One of these was that the instrument should be part of an evaluation system, and should evaluate the specific tasks and objectives that have been identified in the first stages of the learning system. Other principles were that the instrument should be similar to the clinical skill, it should be derived from a content sampling map, it should not be used for two purposes that have conflicting goals, and it should be reliable and valid. Several suggestions were made to increase the reliability of clinical evaluation instruments. The use of comments as a replacement for the measurement of specific objectives and content sampling was discouraged. (Author/BJG)