Abstract: Identifying the most probable point of inverse reliability (MPPIR) is the principle problem in structural inverse reliability analysis. The advanced mean value method (AMV) is widely used in the MPPIR search. However, the AMV method may fail to converge for non-concave and non-convex or high nonlinear performance functions. Based on several improved methods of the AMV, a new MPPIR search algorithm by controlling the search direction and step size during the iteration is proposed in this paper. The calculation procedure combined with inexact one-dimensional search method is also demonstrated. Numerical results show that the proposed algorithm has better convergence compared with the AMV. Further, the proposed algorithm is much easier than the arc search method because the optimization process for step size is not needed.