In previous studies of the agostic bonding in TpMe2NbCl(R′ CCR″)(R), we have made use of a hybrid QM/MM protocol (B3LYP:UFF) where the QM partition ([Nb(Cl)(iPr)(HCCH)-(NHCH2)3] +) was rather small, but the optimized structures were nevertheless in apparently good agreement with experiment. In attempting to improve this model by expanding the size of the QM region, we were surprised to discover that a full QM treatment of the whole molecule using the B3LYP functional failed to locate an agostic structure of any kind. A systematic assessment of density functionate reveals that the poor performance of B3LYP in these systems is typical of all DFT methods that do not obey the uniform electron gas (UEG) correlation limit. Those that do obey the UEG limit, in contrast, provide an excellent description of the agostic structure when the complete ligand system is treated at the QM level. The apparently good performance of our original (B3LYP:UFF) hybrid method can be traced to a cancellation of errors: the B3LYP functional underestimates the intrinsic strength of the agostic interaction relative to competing Nb-Cl π bonding, but this is offset by an additional but unphysical electrostatic component to the agostic bond introduced by the presence of a positive charge in the QM region. © 2007 American Chemical Society.