Tax is a main source of state finance income whose forecasting accuracy plays a significant part in various economical policies making.This paper proposes an improved GM(1,1) model for tax prediction by changing the background values in the traditional GM(1,1) model.Original tax data from 1994 to 2008 are analyzed and forecasting results indicate the grey model can simulate the trend growth embedded in the data quiet well, and the improved GM(1,1) model outperforms the traditional model.The whole process is concise and easy to be programmed, which suggests the proposed forecasting method is feasible for tax forecasting.