Utilization of coupling constants as restraints in computational structure refinement is reviewed. In addition, we address the effect of conformational averaging and examine different approaches to apply the restraints when the experimental observable is obviously a result of averaging. Here, two different computational methods are compared. The simulation of a single structure with time-dependent restraints produces results very similar to those obtained with the calculation of numerous copies of the molecule (an ensemble of structures) and ensemble averaging. The advantages and disadvantages of the two methods are illustrated with simulations of cyclosporin A, for which 117 NOEs and 62 homo- and heteronuclear coupling constants have been measured.