To blindly separate multiple digital band-pass communication signals which are overlapped in time and frequency domains， by using the cuckoo search algorithm， a new separation method is researched based on the independent component analysis. Firstly， the mixed signal model is constructed under uniform linear arrays. Secondly， from the perspective of maximizing the non-Gaussianity of signals， the separation of signals is transformed to the problem of optimizing kurtoses of signals. Finally， the cuckoo search algorithm is used to optimize the cost function which approximates kurtosis. The iteratively generated individuals having the best fitness is regarded as the mix-solving vector and then the source signals are separated. Compared with the existing independent component analysis methods based on fixed-point algorithms， the method in this paper is able to overcome the limitation of its inability to separate the amplitude keying （ASK） signals that are intermixed in the time-frequency domain， and it is applicable to the blind separation of digital bandpass communication signals with any modulation type. The validity of the method is verified and the performance of the method is analyzed through simulation experiments， which show that the separated signals have a high signal-to-inference under the condition of lower signal-to-noise ratio， indicating that the method has high robustness.