Abstract
　The possible extreme of earthquake magnitude in a region is one of the most important factors in assessing the seismic risk. The aim of this paper is to examine whether or not and how accurately we can estimate the extreme value based on the recent instrumental magnitude data which are usually given for rather short period of time.
　The so-called Gutenberg-Richter's formula has been generally used to represent the magnitude distribution of earthquake. Recently, however, a modified distribution function with a truncation on larger magnitude side was introduced. Applying this function to observed data, the possible extreme can be estimated by using the maximum likelihood method. On the other hand, the magnitude of maximum earthquake which actually occurred in that region since historical time is known from seismological catalogs of historical and recent events. The comparison of these two values in many regions gives the answer to the present problem.
　Based on magnitude distributions in 18 regions in Japan obtained from the JMA catalog for 56 years from 1926 to 198l, it is proved that the estimated extremes have a high correlation with the maximum magnitudes in the same region since historical time. The average of differences between two values is almost zero and the standard deviation is about 0.3 in magnitude unit. This result indicates that, if a seismological catalog of recent instrumental data is of uniform quality for sufficiently long period as in Japan, the present method gives a fairly reliable result in assessing the possible maximum earthquake in future.

*東北大学理学部
Faculty of Science, Tohoku University