The paper presents two approaches to recogniser word response modelling which rely on an estimation of interphoneme distances as a basis for modelling recogniser behaviour towards a particular vocabulary. The recogniser specific inter-phoneme distances are derived from testing on series of minimal word pairs in which only a single phoneme varies in fixed phonetic context. In the first approach an existing model of human word recognition is applied and modified to incorporate recogniser specific inter-phoneme distances rather than the original inter-phoneme distances obtained from human perception experiments. Experiments are conducted on both a DTW- and a CDHMM-recogniser architecture and the prediction capabilities are compared with "live" results from these two recognisers. These experiments lead to a formulation of the second approach, a new HMM-based response model.