The satellite technology has been attracting the attention of governments and companies with increasing areas of application. For any of these application areas, reliable and rugged design of satellite is essential. In case there is a failure on satellite, according to its severity, it may lead to the loss of the mission or even the satellite. Therefore, assessing possible satellite failures that might occur during its operational life and taking precautions in the design phase is very critical for satellite programs. In general, numerical reliability analysis starts with setting numerical reliability objective at satellite level, and then allocating the numerical value within lower levels. After all, analysis is verified with performing bottom-up reliability prediction analysis. However, typical reliability prediction analysis uses failure rate data from well-known reliability libraries, statistical evaluations, probability distribution models and all of them include various uncertainties. Because of these uncertainties, reliability prediction result may be too conservative or the opposite way round too loose which can result with overdesign or under-design satellite respectively. In this study, in order to reduce these design handicaps, usage of real life data for reliability prediction analysis will be evaluated. A method to compare the reliability prediction analysis of a satellite sub-system with the results of failure estimation algorithm using the real satellite failures observed in space will be proposed.