The refill records of computerized pharmacy systems are used increasingly as a source of compliance information. We reviewed the English-language literature to develop a typology of methods for assessing refill compliance (RC), to describe the epidemiology of compliance in obtaining medications, to identify studies that attempted to validate RC measures, to describe clinical features that predicted RC, and to describe the uses of RC measures in epidemiologic and health services research. In most of the 41 studies reviewed, patients obtained less medication than prescribed; gaps in treatment were common. Of the studies that assessed the validity of RC measures, most found significant associations between RC and other compliance measures, as well as measures of drug presence (e.g., serum drug levels) or physiologic drug effects. Refill compliance was generally not correlated with demographic characteristics of study populations, was higher among drugs with fewer daily doses, and was inconsistently associated with the total number of drugs prescribed. We conclude that, though some methodologic problems require further study, RC measures can be a useful source of compliance information in population-based studies when direct measurement of medication consumption is not feasible.