Abstract: For the ambiguity and noise of oil log data, only using the rough set in data mining would decrease the classification precision. For the large input information-dimension from database, only using the neural network in data mining would make the structure of neural network complex and the training overtime. To solve above problems, an approach to data mining integrated rough set and neural network is presented in this paper based on oil log interpretation principle. The process follows, log data preprocessing, samples information reduction by rough set, stud-ying-training of neural network, unknown information recognition by neural network, error analysis, and so on. A two-layer neural network with nonlinear connection weights make the operation simple. The examples of lithology recognition and reservoir parameters quantitative calculation show the coincident ratio by the data mining method is much more than that of other single methods, and the effect of log interpretation is satisfactory.