The objective of the history matching process is to build reservoir models consistent with production data as well as geological constraints. A number of automatic history matching techniques have been developed in recent years. Among these methods, gradient-based history matching techniques are becoming more widely used due to their very fast convergence rate. However, they have two major drawbacks: convergence problem for a large number of parameters and local convergence of objective function. In this article, a method to resolve these two drawbacks is proposed. The proposed method takes the advantages of the gradient-based method for fast convergence. Also, it uses Latin Hypercube sampling and Levenberg-Marquardt algorithm to search the global minimum of objective function. To avoid a convergence problem, the number of parameters is reduced by sensitivity and correlation analysis. The method is validated by a case study at which observed oil rate, bottomhole pressure, gas oil ratio, and water cut are matched. The great advantage of such an approach compared to other global optimization methods is to reduce computation time and cost due to using relatively lower simulation runs and a much faster convergence rate to an optimal set of parameters using gradient information.