Like in the "plasticity" theory, the prediction of phase transformation yield surfaces constitutes an essential issue in the modelling of polverystalline shape memory alloys thermornechanical behaviour. Usually for "micro-macro" integration, the nature of the interface between austenite and twinned or untwinned martensite under stress free state and the choice of correspondence variants (CV) or habit plane variant (HPV) are predominant toward the explicit shape of the phase transformation surface. If the predictions for Cu-Al-Be, Cu-AI-Zn (interface between austenite and one single variant of inartensite for cubic to monoclinic phase transformation) and Cu-Al-Ni (interface between austenite and twinned martensite for cubic to orthorhombic phase transformation) are fairly good; the prediction is not efficient in the important case of Ti-Ni (interface between austenite and twinned martensite with stress free state for cubic to monoclinic phase transformation). The usual hypothesis consisting in neglecting the effect of stress on the interface geometrical configuration must be revised. (C) 2004 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.