We reanalysed 16 groundfish surveys at South Georgia to determine whether recruitment data used by the Working Group for toothfish assessments in 1999 and later were correct. We found several unexplainable errors. A new, corrected, series of recruitment estimates was derived using new procedures and 6 strata (3 depth strata each at Shag Rocks and South Georgia). It is easy to spot, visually, the modal progression of peaks in the length densities corresponding to cohorts assessed by sequential surveys. Length based analysis of these confirms the growth rate expected from the CCAMLR growth equation for South Georgia. Although it is easy to spot cohorts, the size of individual cohorts can not be consistently determined from the survey data. The expected reduction in cohort density with time at a log-scale rate of -0.16 (natural mortality) was evident in only very few cohort plots, even those which were particularly obvious as visual modal progressions, and even when the data were trimmed to remove suspect surveys. We conclude that survey-based estimates of recruitment biomass are not reliable input data for assessment models using the GY approach at South Georgia.