Using stomach lavage samples from macaroni penguins Eudyptes chrysolophus Brandt breeding at Bird Island, South Georgia and concurrent net samples caught within the penguin foraging range, we examined the potential selection of different length and maturity stages of Antarctic krill Euphausia superba Dana. Using Monte Carlo randomised simulation techniques, we also determined the probability of obtaining length-frequency distributions of krill different from that obtained in the net samples. The krill taken by the macaroni penguins differed significantly from those caught in the nets. Small krill (28 to 38 mm) were absent from the stomach samples, whereas large krill (58 to 62 mm) were more abundant. Random sampling using Monte Carlo simulation techniques produced length-frequency distributions that were statistically different from the original distribution of krill caught in nets on 76 out of 100 trials. Nevertheless, these differences were smaller than those found between the penguin samples and net samples. Comparison of krill maturity stages showed that krill taken by macaroni penguins contained 3 times as many female as male krill, whereas krill caught in nets contained nearly equal proportions. The differences in size and maturity of krill taken by penguins are discussed in terms of aggregated random sampling, prey selection by predators, and evasion by krill of predators and nets. We conclude that the differences are unlikely to be accounted for simply by sampling anomalies; the differences are more likely to relate to penguins selecting larger, nutritionally superior krill, but might also reflect differential escape responses of particular classes of krill when evading penguins or nets.