Computational studies of amine-boranes (X3C)(m)H3-mB-N(CH3)(n)H3-n (X = H, F; m = 0-3; n = 0-3) show that the B3LYP model performs poorly in predicting the structures and B-N bond dissociation energies of such species. A survey of several models shows that the MP2 approach gives the best agreement, but is too computationally intensive for general use. Among several hybrid and pure DFT approaches, the MPW1K model gives the best agreement with experiment and/or with the MP2 model. Scans of the potential surface for rotation around the B-N bond in several molecules and examinations of other amine-boranes suggest that the difficulty with the B3LYP method does not arise from its inability to incorporate nonbonded intramolecular interactions, but from an inherent inability to model the dative bond. The MPW1K approach evidently does this better because it was designed to model "incompletely bound" transition states, which mimic datively bonded systems.