In the process of OCR image scanning, images are always placed slantwise to some extent. Such skew image may create difficulty to character segmentation and result to lower character recognition accuracy. In order to detect the skew angle of the images and correct them, a skew document image detection method based on Hough transform is presented. To overcome the heavy computing of Hough transform, this method extracts "featured pixels" from the image. Experiment results show that skew angles of many kinds of document images can be efficiently and accurately detected by this method and the method has high adaptability.