Abstract: In this paper，neural network is introduced into the design of predictor，and a neural network predictor for lossless compression of Bayer pattern image is implemented.The predictor which consists of two hidden layers is very suitable for nonlinear prediction，and can be used directly in the lossless compression of CFA raw data.In order to increase the precision of the prediction，an edge detection algorithm is proposed.Experiments show that the algorithm is simple and efficient.