Multi-valued Neuron with Periodic activation function (MVN-P) was proposed for solving classification problems. However, the boundaries between two distinct categories are rigidly specified, resulting in inflexibility and long training time. We propose a revised model, called Multi-valued Neuron with Periodic Tolerant activation function (MVN-PT), in which a zone is provided between two distinct categories. Furthermore, genetic algorithms are applied to automatically decide the optimal size of each zone. As a result, performance can be improved. Simulation results show that MVN-PT can offer a higher classification accuracy and run faster than MVN-P.