As the existing algorithms of blind source separation（BSS） are suitable only for sparse sources but not for incompletely-sparse sources,a statistically non-sparse decomposition principle（SNSDP） is proposed for two observed signals.In this algorithm,the signals are divided into several intervals and the incomplete sparsity of the intervals is determined according to the correlativity of the sources.By utilizing different source recovery strategies for both the sparse and the incompletely-sparse intervals,the estimated sources can be improved.Simulated results indicate that,as compared with the traditional algorithms,the proposed algorithm greatly improves the waveforms of recovered signals and increases the signal-to-interference ratio.