This study proposes the feature selection method to discriminate digitized wheat cultivars. If features of a specific pattern are correlated, the feature selection method can be used to reduce the number of features, thereby eliminating the correlation. The method is based on an orthonormal transformation to find a small set of features that represents samples of a wheat cultivar accurately. In the training stage, produced discriminant functions for bread and durum wheat cultivars yielded nearly 99% and 98% accuracy in the classification, respectively. When the same discriminant functions were used in the test stage, recognition of unknown samples was nearly 82% for durum wheat cultivars and it was about 81% for bread wheat cultivars. (C) 2000 Elsevier Science Ltd. All rights reserved.