A neural network model based on the fuzzy classification concept, called the Connectionist Fuzzy Classifier (CFC), is proposed. The CFC model originates from embedding a "weighted Euclidean distance" fuzzy classification procedure into a four-layered neural network architecture. It employs a one-pass learning algorithm, which can overcome the two major drawbacks of the backpropagation model: the lcocal minimum problem and long training time. Some experiments and comparisons between CFC and some different neural network models are made in this paper. The wxperimental results show that the CFC model has better accuracy for speech recognition than do the PNN, backpropagation, and linear matching methods, especiallyu in a noisy environment.