The sampling procedure commonly used on board research vessels in which catches are sorted by species and picked up into baskets was examined in catches containing several species, principally haddock. The procedure does not appear to permit withdrawal of random samples for lengths or counts of fish. Mixing these basket lots results in sample units which are effectively homogeneous for length composition and reduces the amount of sampling necessary to estimate numbers of fish caught. However, mixing does not appear to be effective if the species sort is omitted, especially in large catches.