The relationship between magnitude and frequency of mega-collapses (i.e., sector collapses), mainly of volcanic edifices, in Japan is examined by using existing datasets for volcanic mega-collapses and smaller but more frequent events. Statistical analysis of these datasets showed that the magnitude-frequency distribution of slope failures with volumes greater than or equal to 10 7 m 3 can be expressed by a simple exponential function: logN(x) = abx, where N(x) is the cumulative number of mass-movement events with magnitude ≥x. When this function was fitted to the datasets, the slope coefficient, b, was 0.7 or 0.8. The frequency distribution of mega-collapses was similar to that of smaller (volume >10 5-6 m 3) events. Records from the past millennium in Japan suggest that this magnitude-frequency relationship may be applicable to the last several tens of thousands of years. Therefore, it is possible to predict event probability and the recurrence interval of events of a certain magnitude. In this way, mega-collapses with a volume of over 10 9 m 3 may be estimated to occur at least every 1000-2000 years somewhere in Japan. Therefore, mega-collapses in Japan should not be considered "rare"; rather, they are "normal" events on a geomorphological timescale.