Calculation of ultimate torsional strength of reinforcement concrete (RC) members due to the lacks of the theory of elasticity is a difficult task. Therefore, the finite element analysis could be applied to determination of strength of concrete beams. As well, for modeling of complicated, highly nonlinear and ambiguous phenomena, artificial neural networks (ANN) are appropriate tools. The main purpose of this paper is an evaluation of ultimate torsional strength of rectangular concrete beams. A three-dimensional finite-element model (FEM) along with establishing the artificial neural network is used for achieving this aim. The finite element model utilizes the brittle failure criterion for concrete fracture, and experimental data are applied for training of the ANN.The commercial software is used for numerical modeling, and existing experimental tests are used invalidation of the proposed failure criterion. In order to apply the data for training of network, they are divided into three categories: training, testing and validating data. For training of the proposed network, three-layer perceptron network with a back propagation error algorithm is used. Comparison of accuracies for applied failure criterion in the numerical modeling, and neural network predictions are done using the experimental tests.
