Abstract: A new maximum entropy stochastic finite element method was proposed on the basis of the dimension-reduction method. In this method, the multi-dimensional random response functions were decomposed into the combination of one-dimensional response functions by the univariate dimension-reduction method, so the multi-dimensional integration which was employed to calculate statistical moments of response of stochastic structures was transformed into one-dimensional integration, and the one-dimensional integration was calculated by the Gauss-Hermite integration. After getting the statistical moments of response of structures, the explicit expression of probability density function of response of structures was obtained using the Maximum Entropy Principle(MEP). The proposed method doesn’t involve the calculation of partial derivatives of response and is fit for nonlinear stochastic problems. The examples illustrate that the proposed method has good accuracy and computational efficiency.