A software test is to execute the program using a test case to examine whether it produces the intended output. For successful regression testing, it is very important to choose a relatively small amount, but productive test cases to maximize testing efficiency. Test case prioritization is a suggested technique for this purpose. This technique arranges the test cases in such a way that higher-order test cases are expected to outperform those on lower-order test cases in fault-finding capability. In this paper, we suggest a new metric for test case prioritization based on the error propagation probability of the test cases. This metric arranges test cases in order by means of the probabilistic fault finding capability of the test cases. Since our metric is based on mathematical probability, it can show statistically consistent and constant results for the fault-finding capability of test cases. The experiment results show that there is a high correlation between the test cases aligned by our metric and their fault-finding capabilities.