The objective of this paper is to analyse some of the most relevant relationships among macroeconomic variables and the agricultural sector in Tunisia. Three alternative models are specified and estimated: a VAR in levels, an unrestricted Vector Error Correction Model (VECM), and a Restricted VECM in which long-run relationships among the relevant variables are identified. In all models short-run dynamics are analysed through the use of Generalised Impulse Response Functions. Results indicate that alternative model specifications generate different short-run dynamics. Long-run identification seems to be a necessary condition for obtaining consistent economic results.