This paper evaluates Brian Zamulinski's argument from considerations of relative likelihood for preferring a ‘religion-as-fiction’ hypothesis to metaphysical realism. The paper finds that the argument fails to consider numerous variant hypotheses, and that the ‘religion-as-fiction’ hypothesis is poorly formulated. It is concluded that an argument from likelihood about the status of religious belief will not, in the way Zamulinski constructs it, give support to a hypothesis unless supplemented by an estimate of its probability. Moreover, once probability is taken into account, the ‘religion-as-fiction’ hypothesis looks very weak.