Abstract: Sonar side scan images are the main data source for identifying oil-bearing sediments. By analyzing the echo characteristics of oil-bearing sediments in sonar images, one can perfom the classification and localization for the sediments and further identify some oil contaminated-areas. The selection, extraction, and classification of image features are key to such algorithms. Based on high-frequency sonar side scan images and expert marking information, this study analyzed the statistical, spectral and grayscale features of the images, constructed an image feature vector library. By using the comination of different features, this sttudy then used the support vector machine algorithm to construct an oil-bearing sediment recognition model, and analyzed and compared the algorithm accuracy under different combinations of feature vectors. The results showed that the Support Vector Machine (SVM) algorithm (based on grayscale features) can recognize oil-bearing sediments in images, with an accuracy rate of over 88%. Finally, we suggested that proposed that the algorithm proposed in this study has highly accuracy and practicability in the identification of oil-bearing sediments, and provides some effective data services and decision support for marine oil spill response.