Abstract: Neural networks are composed of massive parallel processing units. They have unique learning capabilities, which can be used in learning complex nonlinear causal relations, and offering a fundamentally different approach in modeling of constitutive behavior of materials. In this paper, an error-back-propagation (BP) neural network for triaxial constitutive model of concrete was developed, which is suitable to model axial monotonic loading under constant confining pressures. A good agreement between the measured data and the predicted results demonstrates that the BP neural network model with two hidden layers is able to capture significant variability inherent in the concrete samples, and has promising application in structural engineering problems.