This paper compares two approaches to measuring subnational ethnolinguistic diversity in Sub-Saharan Africa, one based on censuses and large-scale population surveys and the other relying on the use of geographic information systems (GIS). The two approaches yield sets of regional fractionalization indices that show a moderately positive correlation, with a stronger association across rural areas. These differences matter for empirical analysis: in a common sample of regions, survey-based indices of deep-rooted diversity show a more strongly negative association with a range of development indicators relative to their highest-quality GIS-based counterparts.