The paper introduces a spatial decomposition of the Gini coefficient that supports the detection of spatial autocorrelation conjointly with an indicator of overall inequality. An additive pairwise decomposition based on a spatial weights matrix partitions inequality between observations that are geographically neighbors and those that are not. A framework for inference on the spatial decomposition is also suggested. The statistical properties of the decomposition measure are evaluated in a Monte Carlo simulation and an empirical illustration involving per capita income inequality in the US states is also provided.