The determination of rock friction at seismic slip rates (about 1 m s(-1)) is of paramount importance in earthquake mechanics, as fault friction controls the stress drop, the mechanical work and the frictional heat generated during slip(1). Given the difficulty in determining friction by seismological methods(1), elucidating constraints are derived from experimental studies(2-9). Here we review a large set of published and unpublished experiments (similar to 300) performed in rotary shear apparatus at slip rates of 0.1-2.6 ms(-1). The experiments indicate a significant decrease in friction (of up to one order of magnitude), which we term fault lubrication, both for cohesive (silicate-built(4-6), quartz-built(3) and carbonate-built(7,8)) rocks and non-cohesive rocks (clay-rich(9), anhydrite, gypsum and dolomite(10) gouges) typical of crustal seismogenic sources. The available mechanical work and the associated temperature rise in the slipping zone trigger(11,12) a number of physicochemical processes (gelification, decarbonation and dehydration reactions, melting and so on) whose products are responsible for fault lubrication. The similarity between (1) experimental and natural fault products and (2) mechanical work measures resulting from these laboratory experiments and seismological estimates(13,14) suggests that it is reasonable to extrapolate experimental data to conditions typical of earthquake nucleation depths (7-15 km). It seems that faults are lubricated during earthquakes, irrespective of the fault rock composition and of the specific weakening mechanism involved.