Abstract: An unsupervised hypersphere neural network that is used for pattern clustering is presented. Each clustered pattern set is a fuzzy set hypersphere with a corresponding membership function. A pattern class is the combination of the pattern sets. The center and radius of the hypersphere are determined using the fuzzy hypersphere learning algorithm, an expansion-contraction process that can learn nonlinear pattern set boundaries in few passed through the data and provides the ability to incorporate new and refine existing pattern sets without retraining. The simulation of clustering demonstrates the superiority of the fuzzy hypersphere clustering neural network.