Analysis of small areas in psychiatric health services research is sensitive to the choice of geographic units. To partially overcome the arbitrary aspects of political boundaries, service areas can be synthesized from empirical data on consumer migration patterns. Using patient flow data from 1991 Medicare discharges, we completed an agglomerative cluster analysis to generate psychiatric health service areas (PHSAs) and found that the optimal solution contained 75 PHSAs. Solutions greater than 75 clusters had high rates of area fragmentation and small single-county clusters. Psychiatric resource supply and health status indicators should be analyzed using expanded geographic units.