A new method of blind source separation (BSS) is proposed in under-determined mixtures based on Gabor transform, and it combines the advantages of Gabor transform and BSS. Firstly, one can obtain the estimate for the number of source signals by the ratios of the coefficients of the mixed signals in Gabor transform field. Then, the appropriate threshold values are selected to process the Gabor transform coefficients, and the inverse Gabor transform is applied to the revised signals. This obtains new combinations of source signals to increase the dimensions of mixed signals. Moreover, the existing BSS methods, such as JADE, can be applied to the new mixed signals with the existing mixed ones, and one can obtain the separated source signals. The effectivity of the method is verified by some simulated signals.