The increasing usage of computer based systems for safety critical operations in applications such as nuclear, space, and automotive systems demands a systematic way of estimating software reliability. The high reliability requirements of safety critical software systems make this task imperative as well. Due to the specifics of software systems and the lack of any universally accepted models it is very difficult to predict the true reliability value of the system. Unfortunately none of the existing software reliability models neither acknowledge nor address this fact. There exist multiple uncertainty factors that influence reliability estimation of safety critical software systems. In this paper, we first define the scope of the important factors in the reliability models and describe a new approach to obtain a realistic estimate for system reliability. For this purpose, we consider different kinds of reliability models also taking into account the system architecture. The influence of uncertainty factors in the models is analyzed to obtain uncertainty bounds. They show an interval, where the true reliability should lie within it. This way system architects may use a so-called worst-case reliability estimation, given by the lower interval bound, for system analysis. We also demonstrate our proposed approach with real data taken from safety-critical applications.