The increasingly drinking water pollution in rural areas have affected people's life. To capture the variation timely and to evaluate the water quality conditions based on monitoring data are of great importance. Based on the rural drinking water source characteristics, 5 major pollutants were selected as the evaluation indexes which included fecal coliform, ammonia nitrogen, total phosphorus, permanganate index and dissolved oxygen, respectively. Projection pursuit evaluation model was established to evaluate water quality standard level and pursuit values of water samples. Then the two projection values would be matched to evaluate the levels of samples. BP neural network was used to study and predict the calculation results and test the rationality 11 surface water quality samples of Dingyuan County in Anhui Province during the wet and the dry seasons were collected for assessement. The results showed that, compared with common assessment methods, this model was of great accuracy and validity, which provided a new way for the comprehensive evaluation of water quality.