This paper examines geographic variations in illiteracy in India, both at the district and state levels, based on 1991 Census data. Multilevel regression modelling techniques are applied in order to establish whether variations relate to the type of people in particular places (compositional effects) or to the characteristics of the places themselves (contextual effects). Results show significant differences between places, which cannot be accounted for solely by compositional effects. These differences are complex and contextual effects are found to be different for different types of people. The analysis provides substantial empirical evidence that policy interventions in the context of illiteracy need to take account of spatial complexity.