There are subjective uncertainty and randomness in concrete carbonation depth forecasting model and the model distribution parameters, which cause significant errors in application to practical engineering. Actual inspection data can not often be used to forecast concrete carbonation depth in the actual project due to its small sample size and lack of sufficient completeness. The weighted value of several model calculations was used to forecast the concrete carbonation depth. By using Bayesian approach, the inspection information and the prior prediction model were incorporated, and the prior model weights and model distribution parameters statistics were updated. It is more accurate to forecast the carbonation depth using the updated model weights and model distribution parameters. The procedure for updating the mechanical model selection and distribution parameter statistics was illustrated with a 10-year-long concrete carbonation test.