This paper integrates finite element (FE) and deep learning (DL) methods to predict the bond stress–slip behavior of reinforcing bars in concrete under static and dynamic loading. A bilinear cohesive model is adopted to describe the bond stress–slip constitutive behavior between the steel reinforcement and the surrounding concrete. After calibrating the cohesive model with published experimental results, the influences of different working conditions on the bond strength between reinforcement and concrete are further investigated by performing a sensitivity analysis of critical cohesive parameters. Regarding their influences on the FE predictions, the cohesive parameters are classified as follows: damage initiation strength, fracture energy, and stiffness. It is found that a higher loading rate increases the bond strength without affecting the trend of the bond stress–displacement response. Furthermore, the FE results are extracted as the training samples for the established NARX DL model. After evaluating the static and dynamic bond stress–displacement responses predicted by the proposed NARX DL model, the Pearson correlation coefficient of 0.97 can be achieved with the FE predictions. Finally, an analytical model suitable for engineering applications is proposed, based on the observed trends, to reveal the dominant mechanisms. Work examples are provided for the loading rates of 0.01 mm/s and 5.00 mm/s. The proposed FE, DL and analytical models provide versatile tools to reasonably estimate the bond stress–slip behavior in reinforced concrete under static and dynamic loadings.