A fundamental problem of the unsupervised classification is often the determination of the valid number of the clusters. This study presents a series of testing procedures to investigate the application of a clustering validity function to the fuzzy c-means (FCM) algorithm. The main objective of the investigation is designed to test the performance of the validity criteria for optimal partition. The testing results from a series of remote sensing images indicate that the validity function indeed can be used as the optimal index for the choice of the cluster numbers for the unsupervised fuzzy c-means classification. Moreover, the tests also provide the proper guide for the determination of the applicable range of the fuzziness index.