The field of cardiovascular disease and atherosclerosis has been enriched by a fruitful interplay between basic and clinical scientists and epidemiologists. Such interaction has been less highly developed in the field of diabetes. I present an epidemiological perspective on certain problems in type II (non-insulin-dependent) diabetes research and point out some potentially useful directions for future interdisciplinary efforts involving both basic and clinical scientists and epidemiologists. Early research on diabetes has been marred by variable case definitions. The demonstration in numerous epidemiological studies that plasma glucose has a bimodal distribution in populations has contributed to the widely accepted National Diabetes Data Group (NDDG) and the nearly identical World Health Organization (WHO) definitions of diabetes. This development has markedly ameliorated many of the difficulties resulting from variable case definitions. Appreciation of the phenomenon of bimodality has also contributed to a better understanding of impaired glucose tolerance. Many studies on the pathogenesis of type II diabetes take the form of case-control (i.e., cross-sectional) studies. The weaknesses of this research design are well known to epidemiologists. Prospective studies, a much stronger research design for inferring causal relationships, are much less common in the field of diabetes than in the field of cardiovascular disease and atherosclerosis. Recently, however, two prospective studies have helped define the role of serum insulin levels and insulin resistance as diabetes risk factors. These studies were carried out in small, relatively isolated populations, and there is a great need to replicate these findings in larger, more representative populations and to further evaluate the relative role of insulin secretion and insulin resistance as diabetes risk factors. Because characterization of these metabolic parameters by modern techniques is complex and expensive, such studies will probably have to be carried out in high-risk populations in which the sample-size requirements are more modest.