A Bayesian model updating and damage identification method based on Markov chain sampling was proposed for damage identification of stochastics structures. Firstly, an objective function based on the measured frequency and mode shapes was established, and then the Markov chain Monte Carlo method with delayed rejection was used for stochastics sampling, and the posterior probabilities of structural parameters in intact and damage states were obtained. By comparing the probability density functions of the identified parameters in the two states, the probability and degree of structural damage were finally obtained. The experimental example of steel cantilever beam indicated that based on the modified stochastics parameters, the damage probability and damage degree of each region of the structure are consistent with the real situation.