In this study, we applied a probabilistic methodology to seismic hazard assessment in metropolitan France. For that purpose we determined an attenuation law adapted to the French context. This law holds for peak ground acceleration on stiff bedrock for earthquakes with local magnitudes between 2.5 and 5.6 recorded in near field (at distances between 3 and 50 km). Geological conditions are taken into account by means of a three-categories classification of lithologies based on a 1/1,000,000 geological map. The seismotectonic zonation consists of areas of diffuse seismicity characterized by a frequency-magnitude distribution. In southeastern France, active faults are considered in a test case and are assumed to follow the characteristic earthquake model. We performed hazard curves for six French cities and maps of peak horizontal ground accelerations expected for return periods of 475, 975, and 1975 years in the country. Sensitivity tests have been performed. The uncertainty introduced by ground-motion variability seems minor compared with that due to the choice of the attenuation law. This study points to the importance of testing internal consistency of the various data and laws used in any seismic hazard analysis (in particular, here the type of magnitude used to predict ground motion). If not, some systematic bias is introduced that may result in systematic errors on peak ground acceleration determination. We also show that the introduction of possibly very large and infrequent events, known only from paleoseismic investigations, may have a dramatic impact on the hazard, especially when long periods of time are considered.