Network alignment is a fundamental problem in several domains that aims at mapping nodes across networks. Here, the authors develop a probabilistic approach that assumes that observed networks are errorful copies from a blueprint. The method samples the distribution of alignments, improving accuracy and enabling potential applications.