Texture present in aeromagnetic anomaly images offers an abundance of useful geological information for discriminating between rock types, but current analysis of such images still relies on tedious, human interpretation. This study is believed to be the first effort to quantitatively assess the performance of texture-based digital image analysis for this geophysical exploration application. We computed several texture measures and determined the best subset using automated feature selection techniques. Pattern classification experiments measured the ability of various texture measures to automatically predict rock types. The classification accuracy was significantly better than a priori probability and prior weights-of-evidence results. The accuracy rates and choice of texture measures that minimize the error rate are reported.