The optical pattern recognition is realized by optical correlation, the traditional optical correlation judge the similarity degree of two images by their correlation peak. The fractional Fourier transform contains the spatial information and the spatial-frequency information simultaneously. The fractional correlation can be used to optical pattern recognition too. To overcome the defect that the traditional optical correlator can not recognize images with small differences, a new method based on the fractional correlation was proposed. The theory of the photoelectric hybrid system for realizing this method is given. The simulated results showed that by adjusting the fractional order p, the correlator can not only recognize nearly identical images efficiently, but also provide the position information of differences. As the fractional order p approach zero, the fractional difference correlation peak became to the difference of two images; as the fractional order p approach one the result is the same as the result of difference correlation.