We compare chemical transport simulations performed in a model using archived meteorological data (an off-line transport model) to those performed in a model in which the meteorological data are predicted every time step (an on-line model). We identify the errors associated with using data sampled at timescales much longer than those operating in the atmosphere or in the on-line model, and strategies for ameliorating those errors. The evaluation is performed in the context of a global off-line chemical transport model called the Model of Atmospheric Transport and Chemistry (MATCH) for three test problems: (1) the passive advection of blobs initially concentrated in the lower and upper troposphere; (2) the surface emission of radon and its decay to lead; and (3) the removal of lead from the atmosphere by wet and dry deposition processes. These problems exercise the important processes of transport by resolved scale winds, rapid transport by smaller scale convection processes, and wet removal (which depends on the representation of the hydrologic cycle). We show that the errors in off-line model simulations (compared to the on-line simulations) can be made small when the sampling interval is order 6 hours or less. We also show that one can accurately reproduce the subgrid-scale processes within the off-line model, rather than needing to archive the results of those processes as input to the off-line model. This suggests that for the spatial and temporal scales treated in global models it is possible to treat many problems nearly as accurately in an off-line mode as one can with an on-line treatment. ¿ 1997 American Geophysical Union