In order to resolve the problem of not taking into account color, texture and shape features in crop disease intelligent recognition systems, feature extraction method based on three feature types was proposed. Two types of color spaces such as RGB and HIS were considered, and the transformation formula of the two color spaces was improved. Then ten color features were defined and extracted. Meanwhile sixteen texture features were defined and extracted based on gray level co-occurrence matrix. And thirteen shape features were defined and extracted based on invariant moment theory. Finally the feature dataset was received which was suitable for identifying four types of wheat leaf diseases. The results show that the system recognition rate is relatively high, and can meet the practical application requirements.