This paper investigates the differences in early occupational earnings of UK male graduates by degree subject during the period 1980-1993. We match administrative student-level data from the Universities' Statistical Record (USR) and occupational earnings information from the New Earnings Survey (NES). The paper estimates relative earnings premia by degree subject using three alternative modeling approaches to control for student self-selection into university courses: i) a proxying and matching method, ii) a propensity score matching method, and iii) a simultaneous equations model of subject choice and earnings determination. Our analysis shows that there is a substantial amount of sample selection originating from unobservable student characteristics. The ranking of university subjects based on relative earnings premia is sensitive not only to the modelling approach but also to time, showing that analyses focusing on single-year data may not generalise to other periods.