This paper studies the loss of human capital that emigration generates in the country of origin. To that end I estimate the human capital distribution of emigrants had they not migrated. Unlike previous studies, I take into account the selection of migrants in terms of unobserved characteristics that affect their productivity. Wages in Mexico of those migrants who come back home after being abroad for some time will be crucial to learn something about the selection of non-returning migrants in terms of unobserved productivity. To test whether returning migrants' wages contain any useful information, I follow two steps. First, I use the model of Borjas and Bratsberg (1986) to show that, regardless of the cause for coming back, the distribution of abilities of non-returning migrants is more similar to the distribution of temporary migrants than to that of non-migrants. Moreover, I test some implications of the model in the data. Second, I show that returning migrants' wages reflect their pre-emigration productivity and are not affected by possible human capital gains derived from the decision to emigrate. Taking into account all this evidence, I use returning migrants' wages in Mexico upon return to estimate the distribution of human capital of non-returning migrants had they not migrated. I show that emigrants come form the middle part of the distribution of human capital in the origin country. I find evidence that taking unobserved human capital factors into account is relevant for the dispersion of the estimated distribution as well as for each of its quantiles. Moreover, it does not greatly affect the aggregate mean of human capital.