This meta‐analysis reviews the intrasector heterogeneity of productivity spillovers from foreign direct investment (FDI) in 31 developing countries through a larger more comprehensive data set. We investigate how the inconsistencies in the reported spillover findings are affected by publication bias, characteristics of the data, estimation techniques, and empirical specification, analyzing 1450 spillover estimates from 69 empirical studies published in 1986–2013. Our findings suggest that reported FDI spillover estimates are affected by publication bias. In combination with model misspecification of the primary studies, the bias overstates the genuine underlying meta‐effect, but the meta‐effect remains economically and statistically significant. Our results emphasize that spillovers and their sign largely depend systematically on specification characteristics of the primary studies and publication bias. Publication bias is not caused by “best practice” choices. Future research needs to cover more developing countries and to investigate not only whether spillovers occur, but also to explore inside the black box of how spillovers actually emerge.