Between 1990 and 1995, we carried out seven Global Positioning System (GPS) campaigns in the Corinth rift area in order to constrain the spatial and temporal crustal deformation of this active zone. The network, 193 points over ∼10,000 km2, samples most of the active faults. In order to estimate the deformation over a longer period, 159 of those points are also Greek triangulation pillars previously measured between 1966 and 1972. Two earthquakes of magnitude 6.2 and 5.9 have occurred in the network since it was installed. The extension rate deduced from the analysis of the different GPS data sets is 14±2 mm/yr oriented N9° in the west, 13±3 mm/yr oriented S-N in the center, and 10±4 mm/yr oriented N19°W in the east of the gulf. The comparison between GPS and triangulation gives higher rates and less angular divergence (25±7 mm/yr, N4°E; 22±7 mm/yr, S-N; 20±7 mm/yr, N15°W, respectively). Both sets of data indicate that the deforming zone is very narrow (10–15 km) in the west, might be wider in the center (15–20 km), and is more diffuse in the east. The analysis of the displacements observed after the Ms = 6.2, June 15, 1995, and the Ms = S.9, November 18, 1992, earthquakes, both located in the west of the gulf, together with seismological and tectonic observations shows that these two earthquakes occurred on low-angle (≤35°) north dipping normal faults located between 4.5 and 10 km depth in the inner part of the rift. Assuming that the deformation is concentrated in relatively narrow deforming zones, we use a simple model of a dislocation in an elastic half-space to study the implication of the localization. Using the geometry of the known seismogenic faults, our observations imply continuous aseismic deformation in the uppermost crust of the inner rift. This model predicts geodetic strain rates close to seismic strain rates in opposition to previous estimates. This is because our model takes into account the activity on low-angle normal faults in the inner rift and an effective seismogenic layer of 6–7 km, about half that usually assumed.