We study the impact of language deficiency on the health production of childhood migrants to Australia. Our identification strategy relies on a quasi-experiment comparing immigrants arriving at different ages and from different linguistic origins by utilising a measure of differences along a continuous range of linguistic distances. Our main results indicate a large negative effect of English deficiency on physical health that is robust to a range of different specifications. In the presence of considerable non-classical measurement error in self-reported language proficiency, our results provide lower and upper bounds for the true effect of English deficiency on health of one half and a full standard deviation in the health score respectively. The empirical analysis is framed in terms of a Grossman model which indicates a twofold role of language skills in health production: language deficiency directly affects the efficiency of health production and indirectly affects access to health inputs. We provide some suggestive evidence on the relative importance of these distinct roles.