Abstract: Decision tree, BP neural network, and logistic model were used to explored farmland classification of Longchuan Country. The effectiveness of results was analyzed. Confusion matrix was adapted to probe into accuracy of the classification. The results showed that the influences of the number of samples were different to three models. With more training samples, BP neural network and decision tree had heavier influence and higher accuracy in comparison with logistic model. Besides of three models, BP neural network had the highest accuracy and needed a longer time to train model with poor interpretation; decision tree had higher accurate and good interpretation; Logistic model performed worst, Therefore, decision tree might be the best model for farmland classification in Longchuan Country. So data mining classification is an effective and exact method for farmland evaluation, which will enhance the precision and accuracy of land evaluation, and is of significance for the optimization of farmland classification method.