On the basis of the stochastic kernel approach, the paper verifies the agricultural convergence process from 1994-2003 across a sample of 170 EU-15 regions at Nuts2 level classified according to their initial competitive profile through the Kohonen Self-Organizing Maps technique. The results achieved have, first, underlined the complexity of the competitive scenario summarized in four different profiles: non competitive regions; strongly competitive and innovative regions; competitive regions with diversification and production sustainability; and strongly dependent on the CAP direct support. They affect in a different way convergence. The process seems to be determined by the specific factors at the basis of agricultural competitiveness within which innovation seems to be the most relevant. On the contrary, persistence prevails across the regions with a relative high dependence on the CAP subsidies. The evidences suggest the need of an agricultural enterprise policy aimed at establishing an environment conductive to farms growth and innovation and designed according to the specific territorial features. This is where the interventions of Regions and Member States come in and the reinforcement of their competitive capacity becomes a pre-condition and an indispensable component of a comprehensive strategy aimed at developing agricultural competitiveness and convergence in the EU regions.