Software industries require early software defect prediction for quality evaluation and resource planning. In early phases of software development life cycle (SDLC) failure data is not available. Therefore, domain expert opinion may play a very important role in estimating the software defect in early phases of SDLC. In this paper, a model is proposed to predict the software defects before testing phase that focuses on the structure of software development process. The metrics of early artifacts of SDLC is used for model development. Model development and experimental part is presented using Bayesian belief network (BBN). Qualitative value of software metrics and expert opinion are the primary part of this approach. To illustrate the applicability and usability of the proposed approach ten real software project data sets have been used. The predicted software defect using different uncertainty levels of domain expert is analysed and validated with actual defect.