Following several years of experimentation with the GPS radio occultation technique, the 6-satellite FORMOSAT-3/COSMIC (F3C) mission was launched mid-2006 and has been collecting data since then. In this paper we present early findings of research performed at Météo-France regarding the use of these data for assimilation in numerical weather prediction. Benefiting from the dense global coverage allowed by F3C refraction-induced observations, we first assess the quality of these data at four levels: bending angle, refractivity, refractivity lapse rate, and temperature.We compare them with calculations from Météo-France numerical weather forecasts. Learning from these various levels of data we devise quality control procedures that rely on the refractivity lapse rate. Applying a recent methodology developed in data assimilation we calculate observation bending angle error variances for our assimilation system. Using these new quality control procedures and observation error estimates we run an assimilation and forecast experiment with Météo-France's operational global 4DVAR data assimilation system used as a reference. Our results indicate a very clear positive impact of the assimilation of F3C bending angle data in the Southern hemisphere for the prediction of geopotential heights and winds. We also observe an improvement in wind forecast skill in the Northern hemisphere, albeit such an improvement is smaller than in the Southern hemisphere.