A clustering method for the analysis of ambulatory morbidity data is presented. This approach reduces spurious variations resulting from idiosyncratic diagnosis labeling and coding habits of physicians and facilitates the analysis of the content of ambulatory medical care through the use of aggregate morbidity data. The clusters provide a tool that allows for the comparison of the content of practice based on different factors such as provider training, practice organization, and patient characteristics. Ninety-two diagnosis clusters were derived using the 1977 and 1978 National Ambulatory Medical Care Survey (NAMCS). These clusters incorporate 86 per cent of all ambulatory visits to office-based physicians in the contiguous United States. The clusters were constructed based on the consensus of a group of clinicians including both generalists, as well as selected subspecialists representing the spectrum of ambulatory medical practice. The diagnosis clusters presented are compatible with the International Classification of Diseases (ICDA-8 and ICD-9-CM) and the International Classifications of Health Problems in Primary Care (ICHPPC and ICHPPC-2). Several applications demonstrating the utility of the method are presented, and directions for future applications are suggested.