Functional traits that define the ecological role of an organism are increasingly being used to determine and predict responses to environmental change. Functional trait analyses of butterflies remain underexplored compared with other taxa, such as plants. Previous works using butterfly functional traits have not comprehensively addressed issues about the quality of trait datasets used and the relative predictive power of different trait types. We compare the consistency of trait descriptions between six widely used trait sources for the British butterfly fauna. We analysed consistency of trait sources using Fleiss's kappa and intraclass correlation coefficient. Principal component analysis was used to produce species ordinations, comparing outputs to examine which trait sets were better at explaining recent species range and abundance changes within the UK. There was a large range in congruence values for specific traits between sources. No single source can be relied upon to produce accurate trait information for British butterflies. Most trait sets are poor predictors of abundance and occurrence changes but are better at predicting current occurrence. An extensive trait set, supplementing biotope‐related traits with explicit resource‐based information recovers more informative ecological classifications and models than those primarily based on life‐history traits or biotope descriptors. Smaller trait sets do, however, recover the specialist–generalist continuum. We conclude that analyses of distribution and abundance changes that rely on traits are highly dependent on trait source and trait type. For butterflies, traits that are based on measures of biotope occupancy should be avoided in explaining changes of abundance and distribution. Including trait information that describes their resource requirements is essential for such analyses. A http://onlinelibrary.wiley.com/doi/10.1111/1365-2435.13082/suppinfo is available for this article.