A new approach of inverse transient analysis, called LDSA, is proposed to detect leaks in pipe networks based on the combination of transient flow simulations and simulated annealing. Through this method, the information of steady-state head distribution is first determined by a pipe network analysis model called PNSA. The downstream valve is suddenly closed to generate a transient event, and the method of characteristics is applied to simulate the transient flow. Then, the optimization technique, simulated annealing (SA), is used to solve the nonlinear problem with a least-squares criterion objective function. A synthetic pipe network system is considered, and two cases are chosen to test the performance of LDSA in numerical experiments. In the case studies, the LDSA demonstrates that the leak locations and orifices in the synthetic network can be accurately identified. The new approach is effective to estimate the leakage and orifice size in these two cases. The results show successful applications and promising features of the proposed method.