The invention discloses a sparse source signal blind separation method based on grid and density clustering. The method comprises the following steps of expanding a mixing model of two sensor signalsto the mixing model of M sensor signals, and carrying out short-time Fourier transform to obtain an instant mixing model of the sensor signals; finding a clustering center as the estimation of the mixing coefficient, discretizing an alpha-delta plane in a uniform division mode, and taking N grids with the highest density as the mixing coefficient; recovering source signals, wherein for each time frequency point in a time frequency domain, values of M source signals are not zero at most, determining which M signals are not zero through utilization of a method for solving the least l<1> norm solution, and solving unique solutions of the source signals. According to the method, the underdetermined blind separation problem under the condition that the source signals are not sparse enough is solved; through application of density and grid clustering, the computing efficiency is effectively improved; and the method has very important significance in the field of blind signal processing and spare component analysis.