Substantial variation in contraceptive prevalence rates (CPRs) and fertility rates (FRs) between community health workers (CHWs) has been documented since the inception of the Matlab family planning program in rural Bangladesh. The coefficients of variation of these indicators for Matlab CHWs were 7% and 26%, respectively, in 1995. To identify the reasons for these performance variations, geographical information system (GIS) approaches were applied to longitudinal and cross-sectional data on 80 CHWs for the period 1980-95. Each observation in the data-set included catchment area-specific characteristics, measures of CHW efficiency in service delivery, and CHW-specific characteristics for one specific year. A one-unit increase in the average age of target women in the catchment area increases the CPR by 9.2%. The CPR increases by 0.2% for each 1% increase in women's literacy and decreases by 0.1% for each 1% increase in the number of Muslim households in the catchment area. An increase of 1 sq. km in the size of the catchment area reduces the CPR by 3%. CHW performance increases with age up to 45 years and then decreases. Similarly, a 1-year increase in the average age of target women reduces the FR by 2%. A 10% increase in the size of the catchment area reduces program performance by increasing the FR by 1.4%. A single geographic barrier to movement increases the FR by 1%. This analysis indicates that CHW performance can be improved significantly by defining catchment areas through use of GIS techniques. Without knowledge of the spatial distribution of population and the physical barriers to movement, allocating a fixed number of clients per CHW may not be the most efficient approach.