There is a growing interest on models able to anticipate farmers' response to agricultural and environmental policy changes. The Positive Mathematical Programming (PMP) method is being extensively used for evaluating the likely impacts of policy interventions. This paper evaluates the capability of three different PMP approaches to forecast changes in cropping patterns due to the 2003 CAP reform: the standard approach (Howitt 1995, Arfini and Paris 1995), the maximum entropy approach (Paris and Howitt 1998) and the Röhm and Dabbert approach (2003). However, neither of those approaches allows for activities non-observed in the base situation. An additional approach is therefore suggested to consider new activities already present in the post-reform situation. These approaches have been tested in an irrigated area of Central Italy. All models have been calibrated to the pre-reform situation and then the 2003 CAP reform has been simulated and model results have been compared with observed cropping patterns. Even if all models calibrate perfectly, response behaviour depends on the selected approach. Compared to the standard approach, the Röhm and Dabbert approach shows a too wide substitution between crops belonging to the same group; and the maximum entropy approach performs better only when prior information is considered. The extended PMP version proposed in this paper depicts a more realistic picture of post reform cropping patterns.