Laser-induced breakdown spectroscopy (LIBS) has demonstrated its high potential in both qualitative analysis and quantitative analysis. LIBS combined with self-organizing mapping (SOM) neural network is applied in oil prospecting industry. Some rock core samples, including ash rock, mudstone, shale, sandstone and dolostone, are automatically classified to lay a foundation for lithology on-line identification. Characteristic variables and principal components, which are obtained by feature extraction from raw spectra, are used as inputs of the SOM neural network, respectively. Classification accuracy is 75% and 86% for the two kinds of inputs, respectively. Particularly, all of ash rocks, sandstone and dolostone are classified correctly when principal components are used as the input. The experimental results indicate that LIBS will be capable of fast identification of lithology after improving the classification accuracy of mudstone and shale.