A shallow M6.4 earthquake with a M5.6 foreshock and the M5.5 largest aftershock occurred in northern Miyagi Prefecture, northeastern Japan on July 26, 2003. The coseismic displacement was observed not only by GEONET (continuous GPS) but also by campaign GPS, leveling, and InSAR. We have plenty of displacement data by multiple geodetic measurements, which is unusually abundant as the M6-class earthquakes. The leveling route in the epicentral area was surveyed on the day just before the earthquake by chance, and that it was resurveyed just after the earthquake. RADARSAT SAR interferograms provide high resolution of surface displacements which reach 240 mm. We construct coseismic rectangular fault model by inversion of the multiple geodetic data. The number of observation points used in the inversion are 23, 17, 49 and 1601 for continuous GPS, campaign GPS, leveling, and InSAR, respectively. We found that two thrust-type fault segments whose strike differs by 50degrees were necessary to reproduce the observed deformation. Seismic moment for both segments is 1.8 x 10(18) N(.)m (M-w 6.1), assuming rigidity of 30 GPa. Because rake angle of two segments is different by only 15degrees, azimuth of slip vectors is significantly different between two faults. Two segments having different orientation can explain the apparent disagreement of focal mechanisms estimated from initial phase of P wave and CMT waveform inversion.