The three stages of academic language testing in the UK over the last 50 years move from grammar through real life to features of language use. At the same time, comparison of predictive validities suggests that all three measures account for very similar shares of the variance (about 10%) and that therefore the choice of an academic language proficiency test is determined only in part by predictive validity: other factors such as test delivery, test renewal in response to fashion, research and impact on stakeholders, and assessment of all four language skills are also important.