It is difficult to separate the overload characteristic signal of projectile penetrating target from complex test signals using the traditional blind source separation (BSS) method. In this study, a new BSS method of penetrating overload signals is proposed, which is not affected by the number of test sensors. This method can also estimate the number of signal sources. Firstly, the singlechannel penetration overload signal is decomposed by the ensemble empirical mode decomposition, and the decomposed intrinsic mode function and the test signal are used to generate multidimensional signals. Secondly, the multidimensional signals are decomposed by the singular value decomposition method. The number of vibration sources is estimated according to the rule of prior Korder singular value dominance. And the maximum crosscorrelation coefficient method is used to determine the best IMF. The test signal and the best IMF are formulated into a multichannel mixed signal. Finally, the multichannel mixed signal is whitened and jointly approximated diagonalzed. The unitary matrix is calculated to obtain the mixed estimation of the original test signal. The acceleration characteristic signal with a correlation of 09747 is obtained by using the method in single channel penetration overload experiment. Compared with the existing methods, this method can effectively separate the characteristic signals of penetration overload. And the adaptive properties of the signal processing process also solves the problem of choosing the filtering frequency of overload signal under different missile target working conditions.