Abstract: We present an algorithm to detect and diagnose the fault with two classes neural networks for the fault in observations of dynamic precise point positioning occurs.At first,the first neural network training samples online is used to improve the reliability of dynamic model.The second neural network can automatically detect failures,position and delete the fault of observations,and the observations processed can further improve the contribution of dynamic model to the result of navigation.In the data of aircraft,the algorithm can detect the fault of observations,improve the percent of diagnosis and control the influences of the fault to the result of navigation.