The original Penn-Jersey housing market model proposed by Herbert and Stevens (1960) and many of its early modifications are deterministic in construct. It is therefore unreasonable to expect such models to adequately account for observed market behaviour. For more realistic and operationally useful models, a probabilistic approach is required. Several probabilistic versions of the short-run (fixed housing supply) Herbert-Stevens model have been published in the literature, the models being derived by employing either entropy-maximising methodology and/or random utility theory. Limited empirical testing of the models has been reported. The aim of this paper is to consolidate the existing models into a general model structure using probabilistic random utility theory. The major departure from existing versions will be the recognition of within dwelling type attribute variability and the extent to which this variability is perceived by locators. This additional source of variability introduces another parameter into the model. When applied to a published empirical example (Senior and Wilson, 1974), the proposed model results in greatly improved empirical performance.