Rock lithology identification plays an irreplaceable guiding role in many aspects, such as exploration and development of oil and gas fields, study of the origin and evolution of the earth, analysis and prediction of geological hazards etc. Therefore, rock identification and classification are very important for geological exploration and analysis. In order to improve the classification accuracy of rocks, a method of rock surface fingerprint analysis and classification based on laser- induced breakdown spectroscopy (LIBS) was proposed. In the experiment, six rock samples were placed on a three- dimensional displacement platform, and different positions of the rock surface were excited by LIBS to obtain the original spectral data. After removing abnormal points, normalization and other pretreatment operations on the collected spectral data, the characteristic spectral lines of five elements (silicon, aluminum, potassium, sodium and magnesium) with large content differences were determined according to the rock mineral composition, and the element fingerprint was obtained. Then, the support vector machine (SVM) was selected as the classifier for classification. The classification model using the spectral mean and the classification model of multi- dimensional fingerprint fusion were established respectively, and the two classification results were compared. The accuracy of traditional classification model based on spectral mean is 59. 4%, while that of multi- dimensional fingerprint fusion model can reach 96.5%. The experimental results show that the element fingerprint shows the element distribution on the rock surface, which can make full use of the heterogeneous structure information of different kinds of rocks, thus greatly improving the classification accuracy of rocks.