Software safety testing focuses on whether safety-critical software under test fulfills corresponding safety specifications. But unfortunately, it is very difficult to test software safety by using traditional operational reliability demonstration testing methods. The main reason is that different failures may have different impacts on safety. Damage size of a failure is a classified measurement of harm severity of the failure under the most adverse circumstances. Harm severity of a failure could be obtained through harm analysis. A classified software safety weight is the probability that failures in a subset of the corresponding damage size do not emerge. Software safety vector is a vector of classified software safety weights.