Aiming at the bond?slip behavior of reinforced concrete, the finite element model of reinforced concrete bond?slip based on cohesion model was constructed by ABAQUS finite element software. The mesh sensitivity and cohesion parameter sensitivity of the simulation model were explored by energy and load?displacement curves. Aiming at the problem of bond strength of reinforced concrete, a nonlinear autoregressive exogenous network (NARX) was developed to predict the load?displacement curve for reinforced concrete by creating 20 sets of data with the variables of bond length, reinforcement diameter, and loading method. The study shows that the mesh size of 6 mm provides an ideal balance between prediction accuracy and computational cost. Based on the sensitivity of the finite element prediction results, the cohesive parameters are in the sequence of damage initiation strength, fracture energy, and stiffness. The NARX with the prediction accuracy of 99.6% is promising to replace time?consuming numerical simulations and experimental works to achieve an efficient and accurate prediction of the bond strength of reinforced concrete. Such an efficient and accurate prediction method provides a novel and convenient methodology of predicting and designing the bond strength of reinforced concrete.