The accuracy of DFT methods for treating NO bonding to heme groups, both with ferric and ferrous iron, is assessed. These systems are shown to be unusually challenging for obtaining accurate binding energies. The hybrid functionals B3LYP and B3PW91 underestimate the bond energy, and the nonhybrid functional BP86 overestimates it as well as predicting the wrong energy ordering for the different spin states of the heme group prior to NO bonding. Large basis set CCSD(T) calculations on model complexes confirm that B3LYP and B3PW91 underestimate the bond strength for NO, by ca. 10 kcal/mol. This is suggested to be due to an underestimate of the medium-range electron correlation associated with metal-ligand pi bonding in this system.