Three studies were employed to further an understanding of a measurement quandary concerning assessment centres (ACs). A common theme associated with ACs is that they do not appear to measure the trait-based variables that they purport to. To compound this mystery, ACs are found to be predictive of outcome criteria; particularly criteria related to promotion. All three studies took varying perspectives on this measurement dilemma. The first study looked at particular traits that were not formally assessed in ACs, and whether these traits explained variance in overall AC ratings. No definitive evidence was found for this notion; however, tacit knowledge appeared to be associated with a small amount of variance in overall AC ratings in one of the samples under scrutiny. The second study looked at the extent to which assessors and candidates understood the models they were assessing and were being assessed under. Neither party appeared to distinguish trait-based, task-based, or other models as being more or less appropriate. While the first and second studies acknowledged some peripheral issues in the AC literature, the third study addressed the fundamental research question. Specifically, the third study investigated whether an alternative to the prevailing trait paradigm was needed.This study compared two models of assessment in a repeated measures design. One model treated the AC data as though they comprised situationally specific behavioural samples. The second model treated the data as though they were indicative of trait-based responses. Using a generalizablity study, both models demonstrated similar psychometric characteristics, although only data treated under the situationally specific model held a conceptual justification. These findings suggest that the situationally specific task-based model presents a more appropriate means by which to treat AC ratings.