Currently, black-box testing methods are effective yet incomplete. Consequently, test engineers may find it necessary to perform ad hoc customisation for each application under test. In this paper, we present procedures for customising black-box methods that model such "error guessing" in a reproducible and reusable way. As a preliminary evaluation, we customise a generalised representation of black-box methods and compare the effectiveness of the resulting test cases with those derived by two existing methods. Our procedures facilitate the development of both domain-specific and novel experimental black-box methods.