Abstract: Abstract: Performance of a horizontal subsurface constructed wetland (HSSF―CW) running for three years was studied. Response curves of the area removal of total phosphorus (TP) to the changes in water temperature were analyzed for different treatment cells. The temporal changes in the area removal of TP in different treatment cells were simulated by the sinusoidal function. Based on the statistical methods of principal component analysis (PCA) and redundancy analysis (RDA), the main environmental factors influencing the removal of TP were selected. Afterwards, the effluent TP concentration (TPo) was simulated and predicted through the artificial neural network (ANN). The results suggested that the area removal of TP was insensitive to water temperature changes when the water temperature was low (<20℃), while great fluctuations combined with an increase of the area removal of TP occurred as the water temperature increased to a higher degree (>20℃). The highest value of area removal TP (3.27 g/(m2?d)) was reached at the temperature of 24.5℃. The relationship between the area removal of TP and the water temperature in different treatment cells was described by the polynomial function, and consequently reasonable accuracy was obtained (R2=0.1082, p=0.000). The variation of area removal of TP in different months was found to be in line with sinusoidal changes (R2=0.231, p=0.000). The area removal of TP with a plateau of 0.397±0.125 g/(m2?d) observed in August was higher than that in autumn. The average area removal of TP was 0.331±0.132 g/(m2?d) in summer. With the method of PCA and RDA, the relationship between the area removal of TP and different environmental factors was analyzed. As a result, the main impact factors including the influent TP concentration (TPi), wastewater temperature (Temp), flow rate (Flow), dissolved oxygen (DO), pH and evapotranspiration (ET) were found, and subsequently selected as the input parameters for ANN modeling. Comparison of the actual and simulated TPo values indicated a certain accuracy of the model in predicting the trend and scale of TPo in the HSSF―CW (R2=0.677-0.800). The results of this research could provide scientific support for the improvement and management of HSSF―CWs.