OBJECTIVE Our purpose is a descriptive analysis of variations in hospital use among small areas of Maryland. METHODS The data are Maryland patient discharge records from acute care hospitals for 1985-1987 and small area population estimates by age, gender, race, and income. RESULTS The common finding was excess geographic variability among Maryland's 115 areas. The hypothesis of uniform rates was rejected for most DRGs, including low-variation mastectomy and hernia repair. Clustering of high-use rates occurred in neighboring areas for orthopedic, vascular, and elective procedures. Admission rates for most nondiscretionary procedures and medical DRGs were reduced in affluent areas while discretionary surgery increased with income level. Elective procedures had extreme variation and were related to income. Coronary artery disease rates declined with income while coronary artery procedure rates increased, indicating that access and patient selection were factors in the use of coronary bypass and angioplasty. CONCLUSIONS The issue is not the ubiquitous variation among small areas but its extent and identification of geographic patterns. Hospital use is related to demography, morbidity, medical resources, access, selection for care, and physician practice patterns. Heterogeneity of these factors ensures that uniform delivery of health care rarely holds. There is little evidence that incidence of surgical disease is the main source of variation in use of discretionary surgery. Rather, variations reflect differing medical opinion on appropriate use. Without evaluation, excessive use cannot be distinguished from underservice. Morbidity explains the variability of nondiscretionary surgery and conditions related to lifestyle. Access plays an important role for discretionary surgery. Geographic analysis can identify variation and relate incidence to socioeconomic and specific local effects. Hospital data do not permit direct assessment of appropriate care. Understanding the reasons for variation requires information beyond incidence data. The challenge is to identify and explain small area variations or to fix them.