This paper introduces our work on how to use image mining techniques to detect SARS, the severe acute respiratory syndrome, automatically as the prototype of computer aided detection/diagnosis (CAD) system. Data used in this paper are digitalized PA(posterior anterior) X-ray images stored in the real-life picture archiving and communication system (PACS) of the 2nd Affiliation Hospital of Guangzhou Medical College. Association rule mining was applied first but results showed there was no significant difference between the locations of the lesions or infiltrate. Classification based on image textures was performed. A sample set contains both the pneumonia and SARS X-ray images was built in the first place. After modeling each sample by a feature vector, the sample set was partitioned to match the detection purpose: classification. Three methods were used: C4.5, neural network (NN) and CART. Final result shows that 70.94% SARS cases can be detected by CART. Data preparation, segmentation, feature extraction and data mining steps, with corresponding techniques are included in this paper. ROC charts and confusion matrix by all three methods are given and analyzed.