Software project development has always been associated with high failure rate. In this paper, we identify the key software risk factors responsible in achieving successful outcome and use a neural network approach to establish a model for minimizing the risks attributed to failed projects. Input of the model is software risk factors that were obtained through interview, and output of the model describes the final outcome of the project. The data for analysis is from real software projects collected through questionnaires. In order to enhance model performance, principal component analysis and genetic algorithm are employed. The experimental result indicates that the software risk analysis can be improved through these methods and that the risk analysis model is effective.