We analyze permanent Global Positioning System (GPS) data obtained over Japan between 1995 and 1997 to estimate the instantaneous interseismic coupling ratio of the seismogenic zones due to the subduction of the Pacific and Philippine Sea plates below the Japanese islands. We first derive the GPS strain rate fields that characterize the crustal deformation of southern and northern Japan and invert them to determine the effective subduction velocity along the central Nankai trough on one side and the Japan-west Kurile trench on the other. These "reference free" velocities are close to those predicted by plate motion models with respect to Eurasia. We conclude that the Eurasian reference frame gives a good approximation to the subduction motion and that to first order, both subduction zones were fully locked during the period of measurements. We then test whether the coupling ratio shows local variations within the seismogenic zones. To do this, we divide the subduction interface into 35 km x 30 km elements that we model by point source groups, and we invert the GPS velocity field referenced to Eurasia to derive the coupling ratio (between 0 and 1) on each fault element. The results are coherent over the 3 years and confirm that both the central Nankai and the Japan-west Kurile seismogenic zones are homogeneously fully locked. Most of the coupling ratios are close to 1 and a few are close to 0; intermediate values are rare. The zones of decoupling correspond either to strong postseismic afterslip associated with the 1994 Sanriku-Oki interplate earthquake (Japan trench) or to a small overestimation of the actual lower limit of the locked zone. We conclude that within the resolution of the GPS data and the model, (1) partial coupling did not exist during these 3 years along the Nankai and Japan-west Kurile trenches; (2) the small seismic coupling ratio previously derived from earthquakes analysis for the Japan and Kurile trenches may indicate that a significant part of the elastic energy is dissipated silently through slow earthquakes and postseismic afterslip; and (3) the heterogeneous coseismic slip pattern observed for the large and great earthquakes that rupture both subduction zones is in great contrast to the homogeneous loading. Finally, we discuss the nonelastic residual deformation within the frame of the long-term deformation of the Japanese islands.