The application of Markov models as deterioration-forecasting tools has been widely documented in the practice of infrastructure management. The Markov chain models employ monitoring data from visual inspection activities over a period of time in order to predict the deterioration progress of infrastructure systems. Monitoring data play a vital part in the managerial framework of infrastructure management. As a matter of course, the accuracy of deterioration prediction and life cycle cost analysis largely depends on the soundness of monitoring data. However, in reality, monitoring data often contain measurement errors and selection biases, which tend to weaken the correctness of estimation results. In this paper, the authors present a hidden Markov model to tackle selection biases in monitoring data. Selection biases are assumed as random variables. Bayesian estimation and Markov Chain Monte Carlo simulation are employed as techniques in tackling the posterior probability distribution, the random generation of condition states, and the model’s parameters. An empirical application to the Japanese national road system is presented to demonstrate the applicability of the model. Estimation results highlight the fact that the properties of the Markov transition matrix have greatly improved in comparison with the properties obtained from applying the conventional multi-stage exponential Markov model.