This paper extracted oil spill which happened in Gulf of Mexico using two images obtained from MODIS on May 17, 2010 and May 24, 2010, and then analyzed the accuracy of extraction results based on the confusion matrix. It was conducted with eigenvectors built by multi-spectral information and spectral matching with the texture characteristics to recognize targets in consideration of mix points phenomenon between oil slick and sea water. Meanwhile, it also made comparisons with minimum distance algorithm and support vector machine to extract oil slick. The contrast among these three algorithms showed that the improved spectral angle mapper (SAM) algorithm had good effect of monitoring the oil spill area than others. Then confusion matrix was used to verify the result accuracy. Results showed that the overall accuracy were up to 90%, and they could meet the extraction accuracy requirement, which could be used in oil spill dynamic monitoring and could provide technical support for oil spill trend and oil sea pollution response.