A new approach is introduced for determining the intrinsic guild classification of a group of species. Previous delimitations of intrinsic guilds have used evidence of spatial distributions (i.e. species co-occurrences), but this is rather indirect evidence. The new method is based on the results of species pairwise competition experiments, and thus uses direct data on species interactions. As with the spatial-distribution intrinsic guild approach, no prior assumptions are made about the classification, nor about which characters are related to guild membership. The method is applied to the results of two published experiments. For one, little independent evidence is available to judge the classification. There is no correlation between the guild classification obtained and gross morphology, but there is no reason to expect any such correlation. For the second experiment, intrinsic guild classifications had previously been obtained from distributional data, and the experimentally-based intrinsic classifications was identical to a distributionally-based one. We suggest that combining evidence from field distributions with experimental evidence offers a rigorous way to determine the true guild structure of communities, offering convincing conclusions when the two lines of evidence converge.