Publication bias (PB) is a major threat to the validity of meta-analysis. Egger’s regression test is the most widely-used tool to assess PB. It can be easily implemented and generally has satisfactory statistical power. It examines the association between the observed effect sizes and their sample standard errors (SEs) among the collected studies; a strong association indicates the presence of PB. However, Egger’s regression may have a seriously inflated false positive rate caused by this association even when no PB appears, particularly in meta-analysis of a binary outcome. Although various alternative methods are available to deal with this problem, they are powerful in specific cases and are less intuitive than Egger’s regression. This article proposes a new approach to assessing PB in meta-analyses of ORs via Bayesian hierarchical models. It reduces false positive rates by using the latent “true” SEs, rather than the observed sample SEs, to perform Egger-type regression; those “true” SEs can be feasibly estimated with Markov chain Monte Carlo algorithm. We present extensive simulations and three case studies to illustrate the performance of the proposed method.