An algorithm for blind recognition of (n,k,m) non-systematic convolutional code with high BER is proposed in information interception. Firstly, an alterable matrix model is constructed to deal with the received data in advance. Then, statistical analysis of the preprocessed data matrixes is carried out to recognize parameters of the convolutional code and extract check-sequences of the each data matrix. Linear equation set using check-sequences is built to figure out the generator polynomial matrix group. Through setting condition of selecting the generator polynomial matrix of the non-systematic convolutional code, the non-systematic convolutional code is recognized. Finally simulation results show that the algorithm can recognize (n,k,m) non-systematic convolutional code effectively in the high BER condition.