This paper asks to what extent host language proficiency can insure immigrants against the risk of ending up in mismatched jobs. Using the 2003-2016 waves of the Household, Income and Labour Dynamics in Australia (HILDA), the paper discriminates between three forms of mismatch, overqualification, under-qualification and over-skilling. Host language proficiency is instrumented using Bleakley and Chin (Rev Econ Stat 86:481–496, 2004) strategy, which exploits the fact that younger children learn languages more easily than older ones. To differentiate between local average treatment effects (LATE) and average treatment effects (ATE), the paper considers two alternative models, 2SLS instrumental variables and biprobit. We find that treatment effects are heterogeneous. English language proficiency among immigrants in Australia reduces the probability of ending up in over-qualified jobs, by between 17.2 (LATE) and 36.7 (ATE) percentage points. The ATE of overs-skilling is also significant and about -8.9 percentage points. In contrast, language skills tend to raise the probability of being under-qualified at the job, by about 8.6 percentage points according to the ATE. Local effects of over-skilling and underqualification fail to be statistically significant, suggesting that host language proficiency may be innocuous for some workers. Overall, the results indicate that host language proficiency is a country-specific, valuable form of human capital.