Crucial to texture classification are texture features and classifiers that operate on the features. There are several approaches to computing texture features. Of particular interest is multichannel filtering because of its simplicity. Multichannel filtering works by decomposing the frequency domain of an image, resulting in a bank of filtered feature images. Many techniques have been proposed to optimize multichannel filtering. However, the optimization is with respect to image representation, thus giving no guarantee for texture classification. This paper proposes a novel technique for learning optimal filters for texture classification. We use regularization techniques such as support vector machines (SVMs) to learn multichannel filters. Since filter training in our approach is naturally tied to classifier training, the resulting filters are optimized for classification. Experimental results validate the efficacy of our proposed technique.