Most formal studies of the clinical reasoning process have been carried out on physicians in the sub-specialties and do not permit generalizations about the nature of the process in the average practising physician. Eighteen family physicians and 19 general internists were randomly selected and assigned to 1 of 4 standardized simulated patient problems in a natural practice setting. Sixty-two physician-patient encounters were studied by direct observation, videotape recall, and subsequent analysis of encounter transcripts by the physician. Physicians consistently developed multiple diagnostic hypotheses early in the patient encounter to guide their inquiry in a manner that would allow them to choose the appropriate hypothesis. Their approach was primarily problem-oriented and was not based on a routine inquiry intended to gather a comprehensive body of data as a basis for making subsequent diagnostic decisions. Little variation occurred between different patient problems. The accuracy and promptness of hypothesis generation were seen to play a significant role in the accuracy of diagnostic formulations. The process must be understood by those concerned with teaching and evaluating clinical competence.