Abstract: Four data mining algorithms are employed to build models, including K-nearest neighbors, naïve Bayes, decision tree and supported vector machine. Based on all a-share market stocks'day trading data which is from April 1, 2015 to March 31, 2016, 10 representative traditional technical analysis parameters were calculated. By selecting appropriate samples, four classifiers were constructed combining with real investment demand and sample data was tested for predicting stock's ups and downs. Results show K-nearest neighbors classifier has higher classification accuracy and supported vector machine has higher sensitivity. On the whole, K-nearest neighbors and supported vector machine are more suitable for real investment.