In this paper, a statistical approach based feature selection method for multilayered feedforward neural network for the classification of wood veneer defects is presented. This method focuses on identifying the superfluous input features by defining a Feature Rejection Criteria (FRC). It is based on an analysis of the intra-class and inter-class variation in the features and their correlation within the same class. The initial neural network design uses seventeen features of the acquired image of the wood veneer as inputs and classifies the veneer as clear wood or one of twelve possible defects (thirteen classes). The revised smaller eleven input neural network results in an improvement in the classification accuracy and time.