In order to solve the prediction of nonlinear distribution water quality data, the traditional statistic method is not suitable for nonlinear distribution data, fractal model N(r) = C/rD is used in this paper. The original data is transformed, the data according well with fractal model is chosen, and the variable dimension fractal model is constructed. For illustration, concentration of dissolved oxygen is utilized to show the feasibility of the fractal theory in solving prediction of nonlinear data. Results show that prediction data is consistent with actual value, from the results, the degree of water pollution will be known. With this probing method, similar nonlinear problem can be dealt with.