According to the identification of radar active deception jamming in tracking mode， by analyzing radar signal and jamming models， we propose a recognition algorithm based on multi-dimensional wavelet features. The signal is sampled and accumulated within the first pulse repetition interval among multiple coherent processing intervals of the target echo to constitute a one-dimensional discrete sequence and the multi-scale wavelet coefficients are extracted using the Mallat algorithm. The shift correlation coefficients of wavelet coefficients at different scales are computed to constitute eigenvector for recognition. Constructing signal samples with different parameters for simulation， the recognition accuracy is still higher than 80% at the signal-to-noise ratio of -5 dB， and the algorithm can effectively distinguish between the range false target and the range gate pull off jamming， which proves the accuracy， versatility and preciseness of the algorithm. Moreover， for the actual situation， we can adjust the dimensions of eigenvector， and select more suitable mother wavelet to further optimize the jamming recognition result.