To obtain the time-dependent reliability of corroded steel beam, accelerated corrosion test and monotonic tensile test were carried out on 9 groups of specimens. 3D data of corrosion depth and the nominal yield strength of corroded steel were measured. Probability model of corrosion depth and nominal yield strength of corroded steels was statistically analyzed, and performance function of the corroded steel beam was established. Then, based on probability density evolution theory, the time-dependent failure probability and reliability index of corroded steel beams were obtained, which were compared with those calculated by Monte Carlo simulation (MCS) method. The results of the study show that the corrosion depth on surface and nominal yield strength of corroded steels obey the normal distribution. With the increase of corrosion time, the mean and standard deviations of the corrosion depth increase. With the increase of weight loss ratio, the mean of nominal yield strength decreases whereas the standard deviation of nominal yield strength increases. Probability density evolution method is closer to calculation results of the MCS, which shows that the method proposed in this paper is effective. The randomness of corrosion depth and strength has great influence on the reliability of corroded steel beams.