Abstract: The automatic recognition of human faces is an active subject in the area of computer pattern recognition, which has a wide range of potential applications. A face recognition method is put forward based on the BP neural network. Also discussed are the problem of feature compression of a face image vector, the problem of determining the number of hidden layer’s neural nodes, the problem of normalization of the input vector, and the problem of initialization of connection weights. Experiments have been conducted for a human face image database of 18 persons with 12 images per person. The recognition results show that compared with the 1 NN classifier, the neural network classifier designed can decrease the error rate efficiently.