This work presents a SVM-based Multiple Classifier System (MCS) for pattern recognition of wheat leaf diseases. The proposed system uses stacked generalization structure to combine the classification decisions obtained from three kinds of support vector machines (SVMs)-based classifiers. And three different feature sets including color features, texture features and shape features are used as training sets for three corresponding classifiers. Firstly, these different feature sets are classified by the classifiers in low-level of MCS to different corresponding mid-level categories, which are partly described by the symptom of crop diseases according to the knowledge of plant pathology. Then the mid-level features are extracted from these mid-categories produced from low-level classifiers. Finally high-level SVMs will be trained and correct errors made by the color, texture and shape SVMs to improve the performance of recognition. Compared with previous classifiers for wheat leaf diseases, the proposed approach can obtains better success rate of recognition.