We challenge the use of traditional measures of ethnic density - e.g., the incidence of an ethnic group on the resident population of a special area - when testing the correlation between stronger ethnic networks and health at birth (i.e., birth weight). Using unique data from ltaly on the main 44 ethnicities residing across almost 4,500 municipalities, we propose more insightful measures, as the distribution of immigrant associations or the incidence of ethnicities sharing the same language. We prove that, once fixed effects for the municipality of residence and the ethnic group are included, the correlation between ethnic density and health at birth is not statistically different from zero. However, ethnic density does channel positive effects on health at birth when a negative shock, as the 2008 Great Recession, struck the labor market. Exploiting a quasi-randomized diffusion of the recession, we find that its average negative impact on immigrant newborns was mitigated by stronger ethnic networks. We show that this can be explained by through sorting of the healthier and more fertile ethnic groups, which experienced also lower levels of in utero selection.