Analyzing design vulnerability for soft errors has become a challenging process in large systems with a large number of memory elements. Error injection in a complex system with a sufficiently large sample of error candidates for reasonable accuracy takes a large amount of time. In this paper we describe RAVEN, a statistical method to estimate the outcomes of a system in the presence of soft errors injected into flip-flops, as well as the vulnerability for each memory element. This method takes advantage of fast local simulations for each error injection, and calculates the probabilities for the system outcomes for every possible soft error in a period of time. Experimental results, on an out-of-order processor with SPECINT2000 workloads, show that RAVEN is an order of magnitude faster compared with traditional error injection while maintaining accuracy.