This study investigates the size distribution of large mass movements in the seismically active Zagros mountain belt (southwest Iran) using a new inventory developed in this project. I have identified and documented 335 large landslides detectable in Google EarthTM images. Geometrical measurements and geological information on landslides were extracted from a digital elevation model with a 30m resolution and 140 geological maps at a 1:100000 scale. Individual landslides have volumes ranging from 104 to 3×1010 m3 and cover areas ranging from 103 to 108 m2. The relationship between landslide volumes (VL) and areas (AL) is well described by a power law of the form VL = ALa where a = 1.49 over five orders of magnitude of AL and seven orders of magnitude of VL. The cumulative volume distribution of the largest slope failures (about 50%) are well describe by an inverse power law of the form NL= 107.3 VL-0.675 for VL = 3.2×107m3. I show that the frequency-size (i.e. volume) distribution is heavy tailed, following a power law for the largest landslides (>107 m3) with a scaling exponent ß = 1.51. Non power law behaviour for smaller landslides is probably an artefact due to the relatively low resolution nature of the data, such that we are essentially missing many small landslides. Comparison of the results with other published datasets shows that the Zagros landslides are truly large and show similar scaling behaviour to that observed in other regions, especially to datasets where landslides are relatively large (deep seated) and occur in relatively resistant materials (e.g. consolidated rocks, as opposed to soil).