An improved MAP-EM algorithm is proposed for Bayesian reconstruction in X-ray CT based upon Markov random field priors and the Poisson data model. The improved algorithm can yield better reconstruction than MAP-EM algorithm, and its convergence is faster. The improved method is verified by applications to computer simulation data and real X-ray CT data from two aluminous tubes scans. Experiments results show this method is effective. Reconstructed slice images of the improved algorithm are accurate and clear.