Software reliability, an issue raised long time ago by software engineering researchers, is still a problem that must be faced nowadays. Our approach is to handle reliability of software products making use of a statistical distribution based model and historical data of similar projects. We analyze the software error metrics of large software projects developed in a multinational company in the last five years, searching for a Rayleigh pattern in the defect rate profile. If we can prove empirically such a pattern exists in certain environment, we have a rapid and powerful theoretical model to support project estimation decisions early in the development life cycle when historical data on similar projects is available.