A new classification method of remote sensing image recognition, called Principal Component-Supervised Classification, is presented. Firstly, by means of principal component analysis, the component images are uncorrelated with each other and explain progressively less of the variance found in the original Landsat Thematic Mapper (TM) data in water area. After analyzing the composition of each component image and its eco-environmental implication according to spectrum features of different water types, the existing water types and their distribution features in the water area are known. Then, the training samples are selected based on the sample water types in PCA images and the classification image is produced following one of the decision rules and programs of supervised methods. This PC-Supervised method, selecting training samples based on the result image of PCA without large-area investigation on the ground or water surface, has thd advantages of unsupervised classification and a partition resolution higher than that of cluster analysis, Furthermore, its distinguishing result, applied to water quality recognition in the northern part of Taihu Lake, shows that the presented water types and their distributions are concordant with the conditions of lake body and environmental factors. So, it is indicated that PC-Supervised classification is an effective and practical method for dynamic analysis of water quality using remote sensing information.