Abstract: This paper established numerical models of steel plates with random pitting damage, and performed nonlinear finite element (FE) analyses on 200 models of the pitted plates. The factors affecting the ultimate strength such as slenderness ratio, aspect ratio and corrosion volume ratio, were studied aiming at exploring their effect on the reduction of structural strength. A BP neural network with three layer structures was constructed and then used to predict the ultimate strength of the pitted plates, taking advantage of its excellent ability to handle the nonlinear problem. The input layer of the BP model contained the three influential parameters mentioned above, while the output layer comprised the FE results. The accuracy of the established BP model was validated against the numerous open data from the literature. Significant variation of ultimate strength arises in the randomly pitted plated structures with the same dimensional parameters (slenderness ratio and aspect ratio) under the same degree of pitting degradation, due to the randomness of the pitting corrosion. The strength reduction is affected by random pitting corrosion, as well as structural size. The constructed BP model has good accuracy to predict the ultimate strength of randomly pitted plates with a relative error no more than 10%.