We propose an optimal-basis texture classification strategy performed in the wavelet packet domain, in order to characterize quality-related information from a set of images. The proposed method enables one to select the discriminative texture in accordance with class information. The proposed methodology has several stages: feature extraction, feature selection, feature reduction, and classification. In the feature extraction stage, we used wavelet energy signatures obtained from wavelet packet transform. In the feature selection stage, two simple optimal-basis methods (top-down and bottom-up searching) were used to select discriminative signatures with high Fisher's criterion values. These approaches improve classification accuracy and reduce the number of features used to classify the quality. Our proposed methodology was applied and validated to classify the surface quality of rolled steel sheets. Using this real-world industrial example, we have experimentally shown that the proposed optimal-basis approach is superior to a full-wavelet-packet-based approach, in terms of classification performance and the number of features used.