Model updating based on Bayesian theory for the prediction of natural frequencies of full prestressed concrete beam has been developed. Morris screening method is employed to study the sensitivity of model parameters and elastic modulus and density of concrete are determined to be updated. Cooperative work between finite element analysis program and Markov chain generating program is realized, and multiple chains with different starting points are designed to obtain the posterior distribution of model parameters. It is found that standard deviations of posterior estimates decrease compared to those of prior distributions. In addition, different starting points are selected to discuss their influence on posterior estimates of model parameters. On reducing the uncertainty of posterior estimate, the difference of each natural frequency used as witness resource is compared.