The rate constant of the NH3 + H --> NH2 + H-2 reaction was calculated using the nonvariational transition-state theory for the temperature range 500-2000 K. The quality of the theoretical barrier height depends on the amount of correlation energy taken into account, on the post-MP4 Corrections, and on the spin decontamination when radicals are involved. The quadratic configuration interaction (QCI) method gives values higher than the perturbational post-MP4 SAC4/Al method. The basis set superposition error and the sources of thermal corrections (theoretical or experimental) have minor importance. We found that the two rate constants using the SAC4/Al and QCI barriers represent an upper and lower bound, respectively, to the available experimental values.