Using a large and novel administrative dataset, this paper investigates variation in returns to different higher education ‘degrees’ (subject-institution combinations) in the United Kingdom. Conditioning on a rich set background characteristics, it finds substantial variation in returns, even within subject, across universities with very similar selectivity levels, suggesting degree choices matter a lot for later-life earnings. Selectivity is weakly related to returns through most of the distribution but strongly positively correlated at the top end. Other than selectivity, returns are poorly correlated with observable degree characteristics, which has implications for student choices and the incentives of universities.