For GPS measurement signal in aircraft experiment is often affected by transmission environment, and interfered with impulsive noise, hereby a SVD combined with wavelet neural network to detect and eliminate the impulsive noise method was proposed. The received GPS data is decomposed by SVD, and the decomposed component is acted as the input of wavelet neural network. Letts criterion is adopted to detect the impulsive noise according to the output residue error of the wavelet neural network. For the detection of the interference points of impulse noise, it can use wavelet network output to replace the measured value, so as to eliminate impulsive noise.