Abstract: The damage detection and assessment algorithm of system identification are based on a parameter estimation. As the random properties of parameters are considered, it is necessary to estimate the variation of parameters and correlation coefficient between parameters. A rational method for prior estimation for covariance matrix of parameters is presented in this paper. For different parameters in an element, the covariance matrix of parameters can be obtained by dividing parameters into a series of statistical independent factors and using the covariance matrix relationship between function and variable. The correlation coefficient of parameters of same type in different elements is estimated by test data or empiric formula.