Tumor diagnosis by X-ray images is a standard way in small animals. But it is difficult to decide tumor region of X-ray images, and veterinarians have to diagnose many X-ray images. Therefore, there is increasing demand for the development of CAD (Computer Aided Diagnosis) system to support veterinarians. We use X-ray images of small animals such as dogs. In this paper, automatic detection of tumor region from X-ray images is studied. We use normalized correlation between original image and template image to emphasize tumor region. This template image is based on feature of tumor. We also use Quoit filter to detect tumor candidate regions. Then, we calculate two feature values, mean of intensity and roundness in these regions, and classify these two feature values into four patterns by K-means clustering to decide true tumor region. As a result, some tumor region can be detected.