In view of the underdetermined situation that the number of sensors is less than that of the source signals, we studied the Undertermined Blind Source Separation (UBSS) problem based on Compressed Sensing (CS). Starting with the mathematical models of UBSS and CS, the source signal that has sparseness property is transformed into the issue of sparse signal reconstruction in CS theory. Then, the signal reconstruction algorithm of CS theory in the Sparco framework was applied in UBSS to construct a two-step CS-UBSS algorithm model. The Restricted Isometry Property (RIP) characteristics of the model was proved in theory. The simulation results prove the feasibility and applicability of the algorithm model for voice signals and image signals, which providing a new way for the solving the UBSS problem, especially in such a case that the reconstruction algorithm in CS theory can be applied directly in the recovery of source signals.