This paper discusses an Improved Linear Fitting Algorithm Based on Stock Time Series (SPLR) .Firstly, the algorithm defines a set of stock trend points and traversals the stock's time series, and finds the stock trend points by extremes method and the investors' experience threshold value. Then it finds the important trend point by the difference of the slope of triangle edge. Finally, connecting these trend points which are found by the above method to present piecewise linear of the stock's time series. This paper's experiment compares with several other fitting algorithms and computes the errors, and the results show: this method is more suitable for fitting the stock and has better timing results, and in large data compression it also can express the stock trend better.