Community detection is a fundamental task in citation network analysis, where one seeks to cluster papers, authors, or journals based on some shared features. This paper aims to identify paper communities from citation patterns that may reflect topic similarities. To this aim we use Bayesian estimation of Stochastic Block Models (SBM) with and without the assortative constraint, highlighting their distinct features. We devise collapsed Gibbs samplers to jointly estimate the number of communities, nodes’ memberships, and connection probabilities. This requires a non-trivial adaptation of existing samplers for undirected networks, suitably modified so as to account for the acyclic nature of citations. An illustration on a synthetic dataset is provided.