Fault localization is a physical and time-consuming activity in the debugging process, especially for the software with large size and high complexity. Existing techniques to locate faults can be classified into two categories: component based and statement based. The former is too coarse to locate the accurate place, while the latter is too fine to contain the computation complexity. This paper proposes a new technique, called double-times-locating (DTL) strategy, to locate software faults. For the first time locating, it abstracts function call graph from the code, builds program spectrum to abstract function traces, and then uses model-based diagnosis (MBD) to sort with probability possible functions candidates that have faults. For the second time locating, it uses DStar to locate faults in the functions. Experimental results show that the proposed technique is more effective than the existing statistics based methods.