This research presents an method to select explanatory variables to develop deterioration models for bridges. The ranking of candidate variables are estimated using the covariance analysis between the condition ratings and inspection data. To determine the most stationary set of explanatory variables, weighting factors associated with the investigated year and ranking are introduced. Yearly inspected bridge data are classified into multiple subsets using explanatory variables and the deterioration model is developed for each. The condition ratings for individual bridges are predicted using the deterioration model. The prediction error comparison shows that the more representative and stationary explanatory variables are selected when weighing factors are considered.