The quality assessment of an urban natural-type lake landscape is influenced by various factors which have complicated and nonlinear relationships with each other. Some of these factors are even random and obscure, of which the inherent relationships are difficult to describe by means of conventional methods. In this study, a three-layer BP Neural Network (BP NN) model was built with thirty design schemes for lake landscapes as the specimens, of which the qualities were assessed based on the six indices including water quality, water quantity, hydrography, vegetation, economy and facilities. The results showed that the initial data were fitted much accurately with the BP NN method, and the predictive results of the specimens were very close to the actual values, with relative errors below 5%. Moreover a higher predictive precision was achieved, which indicated that the BP NN after effective training would achieve higher predictive precision and excellent generalization capability in its application to the quality assessment for urban natural-type lake landscape schemes.