The important feature of the projection pursuit (PP) is that it is one of the multivariate methods able to bypass the “curse of dimensionality”. The aim of PP is to find an interesting or characteristic structure by working in low-dimensional linear projections. PP for regression was originally proposed by Friedman and Stuetzle. In this paper, a neuro-fuzzy approach to the projection pursuit regression (PPR) is proposed for nonparametric regression and nonparametric classification. Our proposed method is based on the membership function and the eigenvector of the covariance matrix to avoid being trapped by a local minimum of the projection indices. The dimensionality of predictors is reduced to one or two in order for good use of human ability of instantaneous pattern discovery. The radial basis function neural network is applied to the function approximation in a projected low-dimensional space. The projection direction is also changed by the adaptive learning (steepest descent) method.