An application of an adaptive neuro-fuzzy inference system (ANFIS) has been investigated for partial discharge (PD) pattern recognition. The proposed classifier was used to discriminate between PD patterns occurring in internal voids. Three different void shapes were considered in this work, namely flat, square and narrow. Initially, the input feature vector used for classification was based on 15 statistical parameters. The discrimination capabilities of each feature were assessed by applying discriminant analysis. This analysis suggested that some of the features possess much higher discriminatory power than the others. As a result, a simplified classifier with reduced feature vector has been obtained. The results demonstrate the importance in identifying and removing redundancy in the input feature vector for reliable PD identification.