Sexual size dimorphism in the Cape petrel Daption capense was analysed to improve methods for sexing live birds in field studies. Samples originated from two geographically distant study populations in the Antarctic: Nelson Island, South Shetland Islands (sample N: 83 M, 89 F), and Ardery and O'Connor Islands, Windmill Islands (sample A: 36 M, 27 F). Using single characters with cut points calculated from sexed individuals, about 59–76% (N; best separator tarsus length) and 61–86% (A; best separator tube length) of birds could be assigned to the correct sex. Combination of characters improved performance of sample-specific discriminant functions to a maximum of 84% in the N-sample (five characters) and 95% in the A-sample (four characters). The most useful character sets as well as ranking of individual characters (their weights) differed between samples. At the same time, cross-testing revealed asymmetry in the applicability of sample-specific functions to the other sample. Hence, the method of a generalized discriminant (Auk110: 492–502, 1993) was implemented and tested to develop a species-specific discriminant for the Cape petrel. For all examined character sets, results of their cross-application to the other sample were improved by using character weights estimated from combined samples instead of weights derived only from the original sample. The generalized function (tube length + 1.05*bill depth at tube + 0.72*culmen length + 0.07*wing length) combined with a cut point calculated from a distribution of discriminant scores (assumed mixture of two normal distributions with possibly unequal variances) is suggested as currently the best compromise for sexing Cape petrels in populations for which specific functions are not yet available: comparison with our best sample-specific functions shows a decrease in accuracy by only 1% to 3%. However, examination of other promising characters (head length, bill depth at gonys, mid-toe length) is recommended to improve further a generalized discriminant. In all cases a larger percentage (N: 87–89%, 38 pairs; A: 100%, 27 pairs) of individuals was correctly sexed by within-pair comparison of discriminant scores than according to the sample-specific cut point. No clear evidence for assortative mating with respect to body size has been found in this study. Available biometric data show little evidence for geographical variation in the Cape petrel, which suggests good applicability of generalized sex-discriminants in other populations.