In this paper, we consider the change request data collected from the system test of a large-scale telecommunication software and analyze the types and causes of failures. And we develop statistical models that incorporate a functional relation between the faults and some software metrics. To this end, we consider three possible regression models including a stepwise regression model and two nonlinear models. Three developed models are evaluated with respect to the predictive quality. We also discuss the advantage of proposed models and the application of our model to a new project.