The ability of density functional models in dealing with polymerization mechanisms has been investigated by comparison with high level post-Hartree-Fock methods. Ethylene insertion and chain transfer reactions have been compared for a model of the active species suggested for the Jordan aluminum catalyst [{R'C(NR")(2))}AIR](+). Conventional density functional approaches (BP86) show a strong bias in favor of chain transfer reactions via hydrogen transfer. The B1LYP model provides improved energy barriers. The aluminum model used strongly favors chain transfer over insertion, this preference being further enhanced by lengthening the growing polymer chain. These findings cast some doubts about currently accepted models of the active species in the Jordan catalyst.