The effects of excess aluminum on water distribution system and human health were mainly attributable to the presences of some aluminum species in drinking water. A prediction model for the concentrations of aluminum species was developed using three-layer front feedback artificial neural network method. Results showed that the reaction rates of both inorganic monomeric aluminum and soluble aluminum varied with reaction time and water quality parameters, such as water temperature, pH, total aluminum, fluoride, phosphate and silicate. Their reaction orders were both three. The reaction kinetic parameters of inorganic monomeric aluminum and soluble aluminum could be predicted effectively applying artificial neural network; the correlation coefficients of k and 1/C0(2) between calculated value and predicted value were both greater than 0.999. Aluminum species prediction results in the drinking water of City M showed that when the concentration of total aluminum was less than 0.05 mg x L(-1), the relative prediction error was large for inorganic monomeric aluminum. When the concentration of total aluminum was above 0.05 mg x L(-1), the model could predict inorganic monomeric aluminum and soluble aluminum concentrations effectively, with relative prediction errors of +/- 15% and +/- 10% respectively.