Abstract: Signal interruption is one of the primary causes for SNR decreasing in the echo spectrum of high frequency surface wave radar HFSWR. Based on the traditional signal recovery AR model method, a new signal recovery method with BP neural network is proposed. The performances of both methods are validated by the experimental results. It is shown that: the prediction error with BP method is less than AR model; and the error rate is less than AR model with the growing prediction step; the power loss of sea echo is smaller using BP recovery method.