This study presents six new water column profiles of the silicon isotopic composition (δ30Si) of dissolved silicon (DSi) from the Atlantic and Indian sectors of the Southern Ocean and a variable depth box model of silica cycling in the mixed layer that was constructed to illuminate the evolution of surface ocean δ30Si over the full course of a year. In keeping with previous observations, δ30Si values ranged from +1.9 to +2.4‰ in the mixed layer (ML), +1.2 to +1.7‰ in Winter Water (WW), and +0.9 to +1.4‰ in Circumpolar Deep Water (CDW). These data also confirmed the occurrence of diminished values for ML δ30Si at low DSi concentrations in early austral autumn on the Kerguelen Plateau. The box model was used to investigate whether these low, post-growing season values of δ30Si were related to input of DSi to the ML from basalt weathering, biogenic silica dissolution (with or without isotopic fractionation), the onset of winter mixing, or some combination of the three. Basalt weathering and fractionation during biogenic silica dissolution could both lower ML δ30Si below what would be expected from the extent of biological uptake of DSi. However, the key driver of the early autumn decrease in δ30Si appears to be the switch from bloom growth (with net removal of DSi and net accumulation of biogenic silica (BSi) biomass) to steady state growth (when slow but continuing production of BSi prevented significant net increase in DSi concentrations with diffusive input of DSi from WW but not decrease in ML δ30Si towards WW values). Model results also indicated that fractionation during dissolution has only a negligible effect on the δ30Si of BSi exported throughout the course of the year. However, seasonal changes in export efficiency (e.g. favouring the export of bloom BSi versus the export of BSi produced during other times of the year) should strongly influence the δ30Si of BSi accumulating in marine sediments. Finally, the choice for the parameterisation of the mixing between the ML and the WW in terms of δ30Si (i.e. constant or allowed to vary with the seasonal migration of the thermocline) is critical to take into account in box model simulations of the silica biogeochemical cycle. Altogether, these results suggest that as a paleoceanographic proxy, δ30Si may more reflect the dominant mode of production of the BSi that is exported (i.e. bloom versus steady state growth) rather than strictly the extent of DSi utilisation by diatoms.