The linear regression models of the physical parameters of structures whose variables are the modal parameters of dominant modes are inferred from the dynamic characteristic equations of structure. Based on these models, Bayesian statistics theory is used to obtain the posterior joint probability density function (PDF) of structural physical parameters, from which the marginal PDF and optimal estimate of the physical parameters are obtained with the Markov Chain Monte Carlo (MCMC) methods. The theoretical analysis and numerical simulation of 5DOF shear building show that the approach presented can not only give the optimal estimate and associated plausibility of the physical parameters using modal data of few primary modes but also get high identification accuracy with relatively exact modal parameters used.