The path delay fault model is the most realistic model for delay faults. Testing all the paths in a circuit achieves 100% delay fault coverage according to traditional path delay fault coverage metrics. These metrics result in unrealistically low fault coverage if only a subset of paths is tested, and the real test quality is not reflected. For example, the traditional path delay fault coverage of any practical test for circuit c6288 is close to 0 because this circuit has an exponential number of paths. In this paper, a statistical and realistic path delay fault coverage metric is presented. Then the quality of several existing test sets (path selection methods) is evaluated in terms of local and global delay faults using this metric, in comparison with the transition fault and traditional path delay fault coverage metrics.