Three large earthquakes occurred within the rupture zone of the 1906 Colombia-Ecuador earthquake (MW = 8.8): in 1942 (MS = 7.9); 1958 (MS = 7.8); and 1979 (MS = 7.7). We compared the size and mechanism of these earthquakes by using long-period surface waves, tsunami data, and macroseismic data. The 1979 event is a thrust event with a seismic moment of 2.9 × 1028 dyne-cm, and represents subduction of the Nazca plate beneath South America. The rupture length and direction are 230 km and N40°E, respectively. Examination of old seismograms indicates that the 1906 event is also a thrust event which ruptured in the northeast direction. The seismic moment estimated from the tsunami data and the size of the rupture zone is 2 × 1029 dyne-cm. The 1942 and 1958 events are much smaller (about to of the 1979 event in the seismic moment) than the 1979 event. We conclude that the sum of the seismic moments of the 1942, 1958, and 1979 events is only of that of the 1906 event despite the fact that the sequence of the 1942, 1958, and 1979 events ruptured approximately the same segment as the 1906 event. This difference could be explained by an asperity model in which the fault zone is held by a discrete distribution of asperities with weak zones in between. The weak zone normally behaves aseismically, but slips abruptly only when it is driven by failure of the asperities. A small earthquake represents failure of one asperity, and the rupture zone is pinned at both ends by adjacent asperities so that the effective width and the amount of slip are relatively small. A great earthquake represents failure of more than one asperity, and consequently involves much larger width and slip.