In this paper, we propose a new Genetic Algorithm(GA)-based Multiuser Detector(MUD), and its performance is evaluated by computer simulation compared to both the optimum MUD and the Hopfield neural network(NN) -based MUD when the near-far problem exists. From the results of comprehensive simulation, it is shown that the proposed MUD in this paper can guarantee a close BER performance compared to both the optimum MUD and the Hopfield NN MUD with a considerable reduced complexity under the near-far condition. Furthermore, a more performance improvement than the Hopfield W MUD can be expected when the near-far problem does not exist.