The collection efficiency of cylindrical sediment traps of common design was assessed in situ for more than 2 y in an offshore shelf regime using the 234Th proxy and simultaneously collected hydrological and biogeochemical parameters. The traps were found to systematically record an undertrapping bias of 234Th by, on average, a factor of three (range 0.8 to 10). Seasonal variations in trapping efficiency were related to the seasonally varying ballasting properties of the settling particle pool. Sediment trap 234Th fluxes agreed within a factor of two of the estimated 234Th export from the overlying surface waters in the winter-spring periods in both 1999 and 2000 and appeared related to a particle pool that included, presumably rapidly sinking, mineral particles and diatom tests acting as ballast. In contrast, discrimination against slowly settling organic-rich aggregates of apparent exudate origin resulted in undertrapping of 234Th by factors of three to ten throughout the summerfall seasons. These data are consistent with hydrodynamic predictions that the collection efficiency of sinking particles is inversely related to their intrinsic settling velocities. Recognition of changing particle composition along a settling velocity spectrum combine with carbon mass balance restrictions to suggest that these 234Th-based collection efficiencies, ranging from 0.1 to 1, may not be directly applicable to "correct" sediment trap fluxes of other components, such as organic carbon. However, the 234Th-derived insights of settling-velocity-related undertrapping may beneficially be taken into account when interpreting trap records both in studies of biogeochemical element fluxes and in phytoplankton ecology