A fundamental challenge in computational drug design is the availability of reliable and validated experimental binding and structural data against which theoretical calculations can be compared. In this work a combination of molecular dynamics (MD) simulations and free energy calculations has been used to analyze the structural and thermodynamic basis of ligand recognition by phenylethanolamine N-methyltransferase (PNMT) in an attempt to resolve uncertainties in the available binding and structural data. PNMT catalyzes the conversion of norepinephrine into epinephrine (adrenaline), and inhibitors of PNMT are of potential therapeutic importance in Alzheimer's and Parkinson's disease. Excellent agreement between the calculated and recently revised relative binding free energies to human PNMT was obtained with the average deviation between the calculated and the experimentally determined values being only 0.8 kJ/mol. In this case, the variation in the experimental data over time is much greater than the uncertainties in the theoretical estimates. The calculations have also enabled the refinement of structure-activity relationships in this system, to understand the basis of enantiomeric selectivity of substitution at position three of tetrahydroisoquinoline and to identify the role of specific structural waters. Finally, the calculations suggest that the preferred binding mode of trans-(1S,2S)-2-amino-1-tetralol is similar to that of its epimer cis-(1R,2S)-2-amino-1-tetralol and that the ligand does not adopt the novel binding mode proposed in the pdb entry 2AN5. The work demonstrates how MD simulations and free energy calculations can be used to resolve uncertainties in experimental binding affinities, binding modes, and other aspects related to X-ray refinement and computational drug design.