This paper estimates the return to education using two alternative instrumental variable estimators: one exploits variation in schooling associated with early smoking behaviour, the other uses the raising of the minimum school leaving age. Each instrument estimates a 'local average treatment effect' and my motivation is to analyse the extent to which these differ and which is more appropriate for drawing conclusions about the return to education in Britain. I implement each instrument on the same data from the British Household Panel Survey, and use the over-identification to test the validity of my instruments. I find that the instrument constructed using early smoking behaviour is valid as well as being strong, and argue that it provides a better estimate of the average effect of additional education, akin to ordinary least squares but corrected for endogeneity. I also exploit the dual sources of exogenous variation in schooling to derive a further IV estimate of the return to schooling. I find the OLS estimate to be considerably downward biased (around 4.6%) compared with the IV estimates of 12.9% (early smoking), 10.2% (RoSLA) and 12.5% (both instruments).