Abstract The computational proof model of Bellare and Rogaway for cryptographic protocol analysis is complemented by providing a formal specification of the actions of the adversary and the protocol entities. This allows a matching model to be used in both a machine-generated analysis and a human-generated computational proof. Using a protocol of Jakobsson and Pointcheval as a case study, it is demonstrated that flaws in the protocol could have been found with this approach, providing evidence that the combination of human and computer analysis can be more effective than either alone. As well as finding the known flaw, previously unknown flaws in the protocol are discovered by the automatic analysis.