Existing software quality assessment models mainly focus on the basic quality characteristics of software systems, and ignore characteristics of customer value and developer organization. Therefore, these models can not evaluate the quality of software systems comprehensively and scientifically. This study characterizes the complex dependence relationships among generalized quality variables by Bayesian network, and constructs a more targeted quantitative model for software quality evaluation. The application shows that the model can comprehensively assess the generalized quality characteristics, make a reasonable evaluation of software quality, and find out the key factors affecting software quality based on the reverse reasoning function of Bayesian network.