Abstract: Stock market trend forecasting is difficult because the stock market is affected by many factors which influence each other.Inspired by phase space reconstruction and visual data mining technique,this paper proposed a visual method to predict stock trend changes based on minute-level data.Research on Shanghai and Shenzhen stock market,eight indexes are selected as stock state variables,and a series of stock state evolution graphs are drawn in terms of cycle and step.After analyzing the correlation between the features of stock state evolution graph and stock price trend changes,variance ratios of centroid and area between two adjacent graphs are taken as criteria to catch stock trend change signals.The experimental results which sourced from 1-minute and 5-minute data of Shanghai composite index and Shenzhen component index from January 2010to September 2011 showed stock state evolution graphs has higher accuracy,lower rate of false positives and false negatives than other stock forecast method.