AbstractRapid or unexpected bridge deterioration can lead to partial collapse, which can subsequently hinder transportation activities and result in economic and human losses. Heavily adopted by the research community, Markov chain-based deterioration models assume that bridge conditions exhibit stationary transitions over time. This assumption requires a significantly large, and often difficult to obtain, number of historical records. As such, Markov chain-based deterioration models have been developed within classical nonlinear optimization frameworks that might result in local optimal solutions. Therefore, to enhance the model capability to simulate the temporal state transition, this study develops a Markovian-based deterioration model embedded within a genetic algorithm (GA) framework—a class of evolutionary computing techniques, to overcome local optimality issues. To demonstrate its applicability, the developed model was applied to a relevant data set of previously rehabilitated and unrehabilitated concrete and steel bridges. The developed GA-Markovian model was able to replicate the actual state probabilities for the unrehabilitated bridges within both the calibration and validation periods. The model performance was slightly lower for the previously rehabilitated bridges due to the inherited nonstationary transition. The model developed in the present study can be used to guide effective rehabilitation and replacement strategies, prioritize available resources, and devise data-driven predictive bridge asset management policies and standards.