The priori knowledge of classes on the ground is an interested task to be incorporated in the training stage for supervised classification. In this paper, an approach to incorporate a priori probabilities in back propagation neural network (BPNN), by way of replicating the training data, consisting of pure pixels belonging to a class according to its proportional area coverage have been implemented. By doing this, the abundant classes are assigned more weights than the other classes in the image. The result of this approach has been compared to the results as obtained from maximum likelihood classifier MLC with a priori probability. An alternative approach to this is to incorporate mixed pixels in the training stage of neural network classification, when sufficient number of pure pixels is not available. The results show that a significant improvement in classification accuracy of the order of 20% can be achieved by implementing the first approach (i.e. replication of training data), whereas in the second approach (i.e. incorporation of mixed pixels in training stage), the accuracy increases only by 14%. Thus, both the approaches result in significant improvement in classification accuracy over MLC. However, the second approach, which does not depend on identification of pure pixels in the image and their replication, appears more attractive to produce meaningful and accurate land cover classification from remote sensing data.