lakes or the open sea within 30 km of their dolines and lateral penetration of brine may be responsible for creating paths for drainage; lateral brine infiltration has been observed several kilometres from open rifts (Thomas, 1975) in the Brunt Ice Shelf, and extensive brine layers covering tens of kilometres observed in the Larsen and Wilkins Ice Shelves (Smith, 1972). Penetration of meltwater to more than 200 m depth in ice shelves has been suggested (personal communication from A. Jenkins) as the explanation for isolated spikes in the <5 18 0 profile of an ice core from the Amery Ice Shelf (presented by Robin (1983)), that are similar to nearsurface <5 18 0 values. Therefore, direct meltwater penetration may account for drainage of the Amery Ice Shelf dolines. There is no evidence of current features on George VI Ice Shelf that could be called dolines, though there are certainly many lakes. George VI Ice Shelf was rather cold er in the 1940s and probably also in the earlier decades of the century than the present day (jones, 1990). Warmer years are mainly a result of mild winters, the summer temperatures being close to O°C. In warmer times, it is likely that much thinner surface covers of ice would form above the melt lakes, not allowing the accommodation of significant pressure from expansion of water during freezing. The water would tend to seep away gently rather than drain catastrophically, and we should expect no roof collapse or a doli ne feature to form. The occurrence of dolines on ice shelves in areas where there are substantial supplies of meltwater strongly supports the theory that dolines are melt lakes that freeze over and then suffer drainage, causing the unsupported roof to collapse. Unlike ice blisters, which usually drain through cracking of the roof of the blister forming icings, dolines probably drain through fractures in their base to the sea beneath the ice shelf.