Based on the time series data from the Aral hydrological station for the period of 1958-2005, the paper re-veals the long-term trend and fractal of the annual runoff process in the mainstream of the Tarim River by using thewavelet analysis method and the fractal theory. The main conclusions are as follows: 1) From a large time scale pointof view, i.e. the time scale of 16 (24) years, the annual runoff basically shows a slightly decreasing trend as a wholefrom 1958 to 2005. If the time scale is reduced to 8 (23) or 4 (22) years, the annual runoff still displays the basic trendas the large time scale, but it has fluctuated more obviously during the period. 2) The correlation dimension for theannual runoff process is 3.4307, non-integral, which indicates that the process has both fractal and chaotic characteris-tics. The correlation dimension is above 3, which means that at least four independent variables are needed to describethe dynamics of the annual runoff process. 3) The Hurst exponent for the first period (1958-1973) is 0.5036, whichequals 0.5 approximately and indicates that the annual runoff process is in chaos. The Hurst exponents for the second(1974-1989) and third (1990-2005) periods are both greater than 0.50, which indicate that the annual runoff processshowed a long-enduring characteristic in the two periods. The Hurst exponent for the period from 1990 to 2005 indi-cates that the annual runoffwill show a slightly increasing trend in the 16 years after 2005.