This paper describes the application of geostatistics to weather index insurance in order to systematically analyze spatial basis risk inherent in index insurance contracts. The notion of spatial autocorrelation is in general overlooked by index insurance practitioners, but has profound implications for the effectiveness of the insurance offered. The analysis shows that it is possible to offer contracts from multiple weather stations to a single farmer, and that doing so will likely reduce the basis risk from a single contract. The two major implications of the paper are 1) that index insurance should be offered in more flexible contracts that allow farmers to hedge their production according to their perceptions of basis risk and their appetites for risk, and 2) the tradeoff between local (yield) correlation and spatial correlation needs to be more carefully considered, as it may even be better to offer contracts with poor yield correlation if they can include more spatial coverage.