Recently a suite of six CPMG relaxation dispersion experiments has been described for quantifying millisecond time-scale exchange processes in proteins. The methodology has been applied to study the folding reaction of a G48M Fyn SH3 domain mutant that exchanges between the native state, and low populated unfolded and intermediate states. A complex non-linear global optimization protocol allows extraction of the kinetics and thermodynamics of the 3-site exchange process from the experimental data, as well as reconstruction of the amide group chemical shifts of the excited states. We show here, through a series of Monte-Carlo simulations on various synthetic data sets, that the 3-site exchange parameters extracted for this system on the basis of (15)N single-quantum (SQ) dispersion profiles exclusively, recorded at a single temperature, are significantly in error. While a temperature dependent (15)N study improves the robustness of extracted parameters, as does a combined analysis of (15)N and (1)H SQ data sets measured at a single temperature, the best agreement is observed in cases where the full complement of six dispersion profiles per residue is analyzed.