Radiation induced soft errors have become an important reliability concern in the sub-nanometer regime. Therefore, it is imperative to devise methods to predict the Soft Error Rates (SER) quickly and accurately in combinational circuits. In this paper, we present a novel technique and a tool to compute the SERs of designs employing hierarchical architectures such as adders and multipliers. The technique uses pre-characterized blocks for current generation and propagation and probability theory to estimate the SER in hierarchical architectures. The analysis results of different hierarchical architectures based on characterization of basic blocks such as muxes, counters and partial product generators using the new technique, are presented in this paper. The run time for most of the designs were in the order of few minutes and we obtain an average speedup of 14084X times over HSPICE and 12.25X times over a contemporary tool SEAT-LA. We have also demonstrated the scalability of our technique for various hierarchical circuits. Our technique can also be extended to any block based architecture.