A comprehensive set of teleseismic waveforms from two South American deep-focus earthquakes of the predigital era, the 1970 Colombia (Mw=8.1) and 1963 Peru-Bolivia (Mw=7.7) events, are inverted for source mechanism, seismic moment, rupture history, and centroid depth. The P and SH wave inversion of the Colombia event confirms previous work, indicating that rupture occurred on a plane that dips steeply west. Rupture direction paralleled the trend of the Wadati-Benioff zone. We decompose the source into subevents, based on a source time function which shows two major moment release pulses separated by ~20 s. The first subevent is located near the initiation point at a depth of ~630 km. The main moment release was located ~70 km to the southeast and ~20 km shallower. Rupture subsequently propagated farther southeast. The source time function has an initial subevent accounting for ~30% of the moment release of the entire event, whereas the long-period centroid moment tensor (CMT) analysis <Russakoff et al., 1997> has the initial subevent yielding ~50%. The high-angle nodal plane rotated ~15¿ clockwise during the rupture, explaining the large compensated linear vector dipole (CLVD) component inferred from CMT solutions. Individual subevents have large CLVD and compressive isotropic components. A full moment tensor inversion of the Colombia and 1994 Bolivia events suggests that the initial subevents might contain a large non-double-couple (NDC) component. For the 1963 Peru-Bolivia event, using P waves, rupture propagated NNW for a distance of ~70 km, parallel to the high-angle nodal plane and the trend of the Wadati-Benioff zone. The focal mechanism changed dramatically after the second subevent, causing a very large NDC component. Both events, together with the 1994 Bolivia earthquake, have a precursor separated in space and time from the main rupture and show rupture velocities varying between 3 and 4 km/s between subevents, with <2.0 km/s on average for the entire event. Low seismic efficiencies and rupture velocities support a highly dissipative, temperature-dependent rupture mechanism for large deep-focus South American earthquakes, compared with events in cold subducting slabs. ¿ 1999 American Geophysical Union