The November 20, 1960, Peru, October 20, 1963, Kurile and June 10, 1975, Kurile earthquakes are classified as tsunami earthquakes based on anomalously large tsunami excitation relative to earthquake magnitude. Long-period surface wave analysis indicates double-couple (faulting) mechanisms for all three events rather than single-force mechanisms indicative of submarine landslides. The earthquakes have shallow depths (< 15 km) and are located near the trench axis and seaward of most other thrust zone events beneath the accretionary prism. Body waveform inversion indicates very shallowly dipping thrust faulting mechanisms for the three events, with dip angles of 6°–8°. Surface wave spectral amplitudes and deconvolution of SH waveforms suggests anomalously long source durations and large seismic moments relative to MS. Specifically, the 1963 Kurile event (MS 7.2) shows a duration of 85 s and a moment of 6.0 × 1027 dyn cm (MW 7.8), the 1975 Kurile event (MS 7.0) shows a duration of 60 s and a moment of 2.0 × 1027 dyn cm (MW 7.5), and the 1960 Peru event (MS 6.75) shows a time function consisting of four subevents with a total duration of 110–130 s and a seismic moment of 3.4 × 1027 dyn cm (MW 7.6). Estimated rupture velocities are about 1 km/s or less, but there is no evidence of unusually low stress drops. The August 1, 1968, Philippines event, previously classified as a tsunami earthquake, shows none of the anomalous source properties, and teleseismic tsunami height measurements are sparse; we do not consider this event a tsunami earthquake. Most of the “anomalous” tsunami excitation results from underestimation of earthquake size by MS due to the long source duration; the tsunami heights are not significantly anomalous relative to seismic moment. The slow nature of these events may result from rupture through the sedimentary rock along the basal decollement of the accretionary prism. Standard scaling laws when adjusted for the slow seismic velocity in the source region show an MW - MS relationship similar to that observed for the tsunami earthquakes and predict MS saturation at about 7.3 rather than 8.0 for typical events.