To assess the safety of software based safety critical systems, we firstly analyzed the differences between reliability and safety, then, introduced a safety model based on three-state Markov model and some safety-related metrics in this paper. For safety critical software it is common to demand that all known faults are removed. Thus an operational test for safety critical software takes the form of a specified number of test cases (or a specified critical mission duration) that must be executed unsafe-failure-free. When the previous test has been early terminated as a result of an unsafe failure, it has been proposed that the further test need to be more stringent (i.e. the number of tests that must be executed unsafe-failure-free should increase). In order to solve the problem, a safety testing method based on critical mission duration and Bayesian testing stopping rules is proposed.