In order to improve the quality of combinational circuit test generation, an effective combinational circuits test generation algorithm based threevalued neural networks［1］ is proposed in this paper. This algorithm does not need propagation and backtracks, but represents the combinational circuits as a bidirectional network of neurons using the threevalued neural networks, and constructs the energy function for the network. A genetic algorithm was used to find the global minimal as the test vectors. The problem of the combinational test generation was formulated to an optimization problem. The experimental results on some standard circuits demonstrate that the proposed algorithm has high fault coverage and short test time.