In this paper, we study physician specialty decisions using several unique data sets which include information on almost all Canadian physicians who practised in Canada between 1989 and 1998. Unlike previous studies, we use a truly exogenous measure of potential income across general and specialty medicine to estimate the effect of income on physicians' specialty choices. Furthermore, our estimation procedure allows us to purge the income-effect estimates of non-pecuniary specialty attributes which may be correlated with higher paying specialties. Understanding the effect of potential income (and other variables) on choices is necessary if the desired mix across generalists and specialists as well as across specialties is to be achieved. Our results show that physicians respond to differences in income when making their specialty decisions.