The present article is an effort to quantify the flexural strength of corroded reinforced concrete (RC) beams. Eight RC beams with dimensions of 150 × 200 × 2200 mm were tested in four-point bending, including six beams corroded in a chloride environment with average corrosion of longitudinal reinforcement in the range of 5% to 15%. The remaining two beams were used as control beams. The ultimate strength, failure mode and particularly the relationship between crack width and corrosion level are analyzed and discussed. A two-dimensional non-linear finite element model has been presented to simulate the steel bar corrosion considering the deterioration of concrete strength and steel-concrete bond stress. The generalised regression neural network (GRNN)-based model for the ultimate flexural strength of corroded RC beams has been established based on 123 experimental data collected from previous studies. The experimental results were used to separately validate the non-linear finite element model and novel GRNN model. Experimental results indicate that if the overall corrosion rate is used, the corrosion crack width can be a function of the percentage of section loss. The GRNN model provides superior predictive results compared to previous empirical methods with the highest correlation coefficient (r), and the smallest normalised root mean square error.