With the development of society and economy, the stock market becomes an important part in the financial markets. Real-time forecast of stock prices is conductive for all parties to grasp social and economic development trend and make favorable decisions. This paper proposes an improved model of predicting KNN stock price based on short-term trends. When predicting the stock price of the current trading day, this model considers the price trend of several previous trading days, and theoretically has better performance in prediction. Analysis of the data from experiments shows that the improved model of KNN performs a better and precise result of prediction than the logistic regression models and traditional KNN model.