Abstract: The DFA method is an efficient method to test the long-range correlations in no n-stationary time series, which will induce bias in small scale when the series have high order trend. From the basic standpoint of long-range correla tion, a modified DFA method is used to solve the drawback. Based on the real da ta of Chinese stock markets, the new method removes the bias and exactly tesitfi es the long-range correlations behaviour in our stock markets.