Time-averaging restraints in molecular dynamics simulations were intro
duced to account for the averaging implicit in spectroscopic data. Spa
ce- or molecule-averaging restraints have been used to overcome the fa
ct that not all molecular conformations can be visited during the fini
te time of a simulation of a single molecule. In this work we address
the issue of using the correct Boltzmann weighting for each member of
an ensemble, both in time and in space. It is shown that the molecular
- or space-averaging method is simple in theory, but requires a priori
knowledge of the behaviour of a system. This is illustrated using a f
ive-atom model system and the small cyclic peptide analogue somatostat
in. When different molecular conformers that are separated by energy b
arriers insurmountable on the time scale of a simulation contribute si
gnificantly to a measured NOE intensity, the use of space- or molecule
-averaged distance restraints yields a more appropriate description of
the measured data then conventional single-molecule refinement with o
r without application of time averaging.