Abstract: The Markov chain Monte Carlo(MCMC) method combined with Bayesian theory can not only use prior information flexibly, but also give the uncertainty of solution. There is a bright application prospect in quantitative diagnosis of flash radiography. For the optical path length data of the flash radiographic images, a posterior probability model is built by Bayesian hierarchical model, and the Markov chain is dynamically constructed by Gibbs sampling. Then the statistical results of linear attenuation coefficients and their uncertainty are obtained and compared with the constrained conjugate gradient (CCG) method. The results of numerical experiments show that the reconstruction result of MCMC method is approximately the same as the true data for the ideal FTO optical path length image. In the case of blurring and noise, the reconstructed result is equivalent to that by the CCG method. Even when the blurred optical path length data is interfered by high noise, the result of MCMC is slightly better than that of CCG. More importantly, the uncertainty of the reconstruction can be provided by MCMC method. The related work in this paper verifies the feasibility of MCMC reconstruction of flash radiographic images and lays a good foundation for MCMC reconstruction with blurred and noised transmissivity images.