Abstract: According to the optimization theory and the neural network (NN) theory,an optimum multiuser detector (MUD) is proposed.The optimum MUD problem is taken as the combinatorial optimum problem.Using the neural network which has the ability of fast optimization computing,the optimization neural network MUD is derived. Theoretical analysis and numerical results show that in the aspect of biterror rate and multiple-access-interference, the optimum MUD is better than the conventional and decorrelated MUD, in aspect of "near-far" resistance, the optimum MUD is better than the conventional MUD, and is worse than the decorrelated detector, and the optimum MUD can be easily implemented by VLSI technology.