Abstract: In view of the problems of huge computations, incapacity to deal with large scale reservoirs and unqualified computational accuracy of the current sensitivity analysis methods in numerical reservoir simulation, this paper proposes an adjoint model-based parameter sensitivity analysis method for history matching. Based on the adjoint system theory, an adjoint model with the adjoint variables independent of simulated variables is established and direct resolution of gradient equation can be avoided; coefficient matrices of the adjoint model are constructed according to the state equation of the reservoir percolation model and solving methods of the numerical reservoir simulator. Adjoint variables are acquired by solving the adjoint equations and sensitive coefficient calculation equations are established. The sensitive coefficient matrices of objective functions with respect to control variables are solved by using the adjoint variables. Compared with the commonly used gradient simulator methods and experimental design methods, the advantages of this method are as follows: the coefficient matrices of the adjoint model can be directly derived from the results of the state equation; the amounts of calculations to solve the gradient equation only depend on the quantity of the observation data rather than the number of model parameters. In order to obtain the derivatives of particular observation data with regard to all model parameters, it only needs to construct and solve one corresponding adjoint equation. With high calculation ability; the sensitive coefficients of all producing time for each grid parameter could be acquired simply by forward modeling an original model and inverting an adjoint model only once, thus the efficiency of parameter sensitivity analysis is greatly improved