In this work, a new methodology based on artificial neural networks (ANN) and indexing techniques is used with the aim to improve memory requirements for storing multisource or multispectral remote sensing (MRS) data and at the same time increase classification speed. This methodology features: a) data quantization using a self-organizing map, b) training set reduction to speed up ANN training, c) fast clustering of prototypes, and d) fast indexed classification. Results obtained for both supervised and unsupervised classification to ground-cover categories using, at no loss of generality, a Landsat TM image, show savings in time and memory without a significant compromise of classification performance.