Data collected as part of the CCAMLR Scheme of International Scientific Observation from conventional krill trawl vessels fishing in South Georgia (CCAMLR Subarea 48.3) were analysed using variance components analysis (VCA) to determine the relationship between observer coverage and our ability to reliably estimate parameters of interest, in this case the mean (or median) krill length and the rate of larval fish by-catch. A partial coverage sampling program has been implemented in South Georgia since 2002: observers have been placed on approximately 50% of vessels fishing in any one season, and have been present on board for about 30% of that season. They have achieved a rate of krill sampling equivalent to 18% of total hauls per vessel per season and 31% per vessel per season for larval fish. Between-vessel and between-haul variance was estimated. For krill mean length and larval fish by-catch rate, between-vessel variance (about 45% of total variance) was slightly lower than between-haul variance. However, the ratio is sufficiently close to 50% that sampling needs to be efficient at both vessel and haul level. It is proposed that an efficient sampling proportion, at least for South Georgia, should be >50% of vessels sampled each season, 20% of total season hauls sampled for krill and an equivalent or higher sampling proportion for larval fish. The Scientific Committee’s method of systematic partial coverage appears to have been sufficient in South Georgia to determine appropriate coverage levels in that fishery. The South Georgia data suggest that such strategies should be pursued for at least four years before the Scientific Committee will have sufficient data to determine appropriate sampling strategies in a particular area.