Most successful studies on water quality monitoring by remote sensing mainly focused on lakes, reservoirs, estuaries, bays and oceans. Since neural networks have been widely applied to the nonlinear problem, therefore remotely monitoring river pollution indexes (RPI) using SPOT images could be a practical way. For atmospheric correction procedure Dark Object Subtraction was selected and followed by separating the samples into two groups for the seasonal variation. In order to consider the sampling difficulty on SPOT images with its limited pixel resolution, an unsupervised pre-classification with manual stream water sampling procedure were used for extracting the reliable water pixels from SPOT images. The study adopted artificial neural network (ANN) to examine and compare the predicting results of river pollution index in accordance with samples clustered by ungrouping and seasonal grouping. The results showed that if we grouped the samples in accordance with seasons, it would improve the accuracy of results to more than 90%. Overall accuracy of grouping simulation can still reach acceptable accuracy around 70% for smaller hidden nodes.