The prediction of steel bar corrosion depth is the basis for evaluating the service performance of RC structures. To establish a prediction model for steel bar corrosion depth in RC members under general atmospheric environment, the main parameters affecting the corrosion depth and the influence law were analyzed in this paper. Then, a numerical model and a RBF neural network prediction model were established based on the measured data, and the parameter sensitivity analysis was carried out. Results show that: compared with the numerical model, the RBF neural network has higher efficiency and accuracy in predicting the corrosion depth of steel bars; it can effectively map the complex nonlinear relationship between the influencing parameters and the corrosion depth of steel bars. The results of parameter sensitivity analysis show that the expansive crack width on RC surface has the greatest influence on the corrosion depth of steel bars, followed by other factors, i.e., the diameter of steel bar, the ratio of concrete cover thickness to steel bar diameter, and the compressive strength of concrete. The proposed model can be used to predict the corrosion degree of steel bars and evaluate the remaining service life of RC structures in engineering detection.