Abstract: This paper put forward a kind of fast search strategy applicable to the computer screen image. Firstly, the moving range of the window containing the target image is determined, and the relative coordinates of the target image are calculated to reduce the search space. In addition, the space coordinates of known matching points are statistically classified, and the search priority is defined according to appeared probability of the target point, to build a learningstyle search algorithm. The test results show that, for the computer screen image, the algorithm has significant advantages in computing speed and matching accuracy. This method has been successfully applied in the national largescale scientific instruments work condition monitoring system.