A new three-layer pattern classification neural network_Adaptive Fuzzy Classification Neural Network (AFC) is introduced. AFC allows hidden neurons to grow to meet the demands of the problem during training in order to overcome the difficulty in predetermining the number of neurons. In AFC, each hidden neuron is regarded as a fuzzy cluster representing a shape of hypersphere in n-dimensional pattern space. Furthermore, in AFC's learning algorithm, neuron merge and fission operations are proposed which are quite useful in reducing redundant neurons. Effectiveness of AFC is shown by comparing the simulation results with conventional neural networks through experiments.