A multiple time-window inversion of 53 high-sampling tsunami waveforms on ocean bottom pressure, GPS, coastal wave, and tide gauges shows a temporal and spatial slip distribution during the 2011 Tohoku earthquake. The fault rupture started near the hypocenter and propagated into both deep and shallow parts of the plate interface. Very large, approximately 25 m, slip off Miyagi on the deep part, at a location similar to the previous 869 Jogan earthquake model, was responsible for the initial rise of tsunami waveforms and the recorded tsunami inundation in Sendai and Ishinomaki plains. Huge slip, up to 69 m, occurred on the shallow part near the trench axis 3 min after the rupture initiation. This delayed shallow rupture extended for 400 km with more than 10 m slip, at a location similar to the 1896 Sanriku tsunami earthquake, and was responsible for the peak amplitudes of the tsunami waveforms and the maximum tsunami heights measured on the northern Sanriku coast, 100 km north of the largest slip. The average slip on the entire fault is 9.5 m and the total seismic moment is 4.2 × 10 22 Nm (Mw = 9.0). The large horizontal displacement of seafloor slope is responsible for 20 to 40 % of tsunami amplitudes. The 2011 deep slip alone can reproduce the distribution of the 869 tsunami deposits, indicating that the 869 Jogan earthquake source could be similar to the 2011 earthquake at least in the deep plate interface. The large tsunami at the Fukushima nuclear power station is due to the combination of the deep and shallow slip, or triggering of shallow slip by the deep slip, which was not accounted in the previous tsunami hazard assessments.