This article puts into question the use of the gravity equation to analyze Regional Trade Agreements (RTAs) in Africa. By surveying the field qualitatively and quantitatively (via a meta-analysis) and by leading our own estimations (with bilateral fixed effects, exporter-time and importer-time effects) on different trade flow databases (UN COMTRADE, DOTS and BACI), we find that the RTAs elasticity of trade in Africa are unreliable due to their unrealistic high level. By introducing intranational trade and bilateral trends into the regression specification, we show that the coefficient of RTAs in Africa are either not significant or drastically reduced. Only COMESA is still significant. We then use a simple general equilibrium model to compare the results obtained with these new elasticities regarding the terms of trade and welfare for members of the COMESA. We find strong trade creation effects that are largely compensated by trade diversion. The welfare gain of COMESA is for most members very low (less than 0.2% of growth).