The western Aleutian trench zone is characterized by rapid motion of Pacific plate (PAC) along the North-American plate (NAM) in North-West direction. The trench terminates south-east of Kamchatsky cape where it intersects Kamchatsky trench making a cusp. The whole 2,200-km Aleutian megathrust demonstrates a rapid change in the sense of relative motion of the Pacific and North American plates, from nearly trench normal at Alaska to nearly trench parallel at Kamchatka. This change is accompanied by an increase in the relative plate velocity from 61 to 76 mm/yr (an estimate based on our global GPS solution). Consideration of the regional seismicity shows that the azimuths of the slip vectors for thrust earthquakes along the arc support an idea of strain partitioning, i.e., the motion in seismic ruptures tends to be less oblique than the plate motion [MacCaffrey, 1992]. The strain partitioning predicts a steady westward motion of slivers of the hanging wall along strike-slip faults, resulting in an active collision of the far western arc with Kamchatka [Geist and Scholl, 1994]. We show, using the constrained nonlinear inversion, that the high GPS velocity of Bering Is. can be alternatively explained by elastic strain accumulation resulting from locking at the subduction interface. In this scenario, there is no steady westward drift of arc slivers since the elastic strain is periodically released in earthquakes, with the islands returning to their original positions. Moreover this approach allows to explain specific azimuths and values of GPS velocities in Kamchatka at the cusp by superposed elastic strains. Unless longer time series at larger number of islands are observed, a unique interpretation of GPS velocities on the Aleutians is not possible. Recently, the first epoch of GPS at Medny (Copper) Is. was carried out, in 100 km from Bering and at greater distance from the Aleutian Trench. Comparison of GPS velocities at Bering and Medny should provide evidence on how significant is the strain partitioning.