As the MCP is NP-hard, an efficient approach to treating this problem is to design appropriate recurrent neural networks. We develop a new algorithm for the MCP, which can, to a certain extent, prevent the associated neural network from falling into local optimal points. The proposed algorithm incorporates nonlinear self-feedback into the SLDN algorithm and has distinguished dynamical characteristics. Simulation results show that the performance of proposed algorithm is statistically superior to the SLDN algorithm.