Model-Based diagnosis is an active branch of Artificial Intelligent. The method is a NP-Hard problem, resolving minimal hitting sets from minimal conflict sets. A compounded genetic and simulated annealing algorithm is put forward by mapping hitting sets problem to 0/1 integer programming problem. After providing the genetic simulated annealing (GSA) algorithm, the efficiency and accuracy of GSA algorithm is tested and compared. The GSA algorithm is not only far more efficient than the traditional one, but also can save 1/3 to 1/2 time than the GA algorithm when the number of conflict sets is more than 35. It can get 98% to 100% minimal diagnosis in most conditions.