As generative parameter learning method uses the maximum likelihood as the target，its classification accuracy is low. So, we proposed an efficient discriminative parameter learning algorithm, which uses the classification accuracy as the target. It learns parameters by discriminatively computing the frequencies of parameters from data set. Empirical studies show that this algorithm integrates the advantages of both generative and discriminative learning and it performs as well as the state-of-art classification method SMV, but is significantly more efficient. At last, this method is used in the problem to identify oil and water layers. The accuracy of the conclusion has very important value for oil field development and production.