Downy mildew is one of the most common diseases suffered by greenhouse cucumbers, which may decrease the quality of greenhouse cucumbers and cause great economical loss to the farmers. In order to increase the accuracy of downy mildew diagnosis for greenhouse cucumbers, a downy mildew diagnosis system for greenhouse cucumbers was designed based on image processing. Focusing on the disease spots images captured in greenhouse field, the conditional random fields (CRF) based on segmentation method was utilized for the system to achieve disease spots images. When building the CRF model, decision tree model was used to extend unary potential function, which could effectively improve the accuracy of segmentation. The post-segmentation images and the disease spots images were transferred to HSV color space, and then 25 features, including color, texture and morphology features, were extracted. A subset of features was generated by rough set method. Finally, the RBF based SVM was used for the system to identify the greenhouse cucumber downy mildew. Taking cucumber downy mildew images obtained in greenhouse from agricultural innovation base of institute of plant protection, Tianjin academy of agricultural sciences as an example, the system was tested. The results showed that the segmentation method used by the system could effectively segment the disease spots images, which managed to overcome the noise caused by the illumination and complex background. A subset of 12 features was obtained by rough set method from the original feature set of 25 features, which improved the efficiency of the system. The identification accuracy of cucumber downy mildew reached 90%, which indicated that the downy mildew diagnosis system could meet the requirement of identification for greenhouse cucumbers.