As an essential component of the Kinect sensor, depth image always contains inevitable and unpredictable shadow noises which limit their usability in many 3D vision applications. Therefore, a shadow detection method based on fractional differential was proposed for depth image. Based on the study of Tiansi template defined by fractional differential, a non-linear stretching operator was designed and implemented. This operator can enhance the boundaries information of shadow regions significantly and accomplish the shadow detection effectively at 0.6 order. Compared with other traditional methods, the proposed method can reach up to 0.971 for F-measure, while the other methods were all less than 0.7. The experimental results indicate that the new method can detect shadow noises effectively.