Based on the Armijo-Goldstein and Wolfe-Powell criterions, a new class of inexact one-dimensional linear search method is presented in this paper, although the traditional method works well. In each iterative process, the method produces its acceptable step-length more reasonable. Compared to the common inexact one-dimensional search, the search method gives in this paper can obtain a satisfying descent quantity of the objective function, and have the global convergence as well.