We propose a new method to investigate stress homogeneity along plate boundaries based on the cumulative misfit of individual fault plane solutions, calculated using assumed stress tensors. Using this method, some segments of faults can be defined, without the time-consuming inversions for stress directions from earthquake fault plane solutions. We assume that the misfits are relatively constant within segments of uniform stress orientation but that they change abruptly at boundaries of segments. This assumption is supported by the pattern of misfits observed for about 400 earthquakes located along the Aleutian subduction zone from 170¿E to 145¿W, which occurred between 1964 and 1993. The efficacy of the approach is confirmed by stress inversion analysis of moving windows. The cumulative misfit as a function of strike along the Aleutian arc shows there are five first-order boundaries where the stress directions change significantly. The first one is near 177.5 ¿E, which corresponds to the northward projection of the Rat fracture zone. The second one falls on the Adak fracture zone. The third is near the locus where the Amlia fracture zone intersects the trench. The fourth lies on the easternmost end of the 1957 aftershock zone. The last is located where the two major asperities of the 1964 rupture are separated. The segmentation boundaries found in our analysis may be controlled by the fracture zones in the subducted plate. This observation may be interpreted as due to decoupling within the underthrusting plate along the zones of weakness (the fracture zones) across which stress may not be transmitted fully. The boundaries defined by our method may also be related to the asperity and aftershock distributions of great earthquakes. The focal mechanism data in the segments defined as stress-homogeneous by our method are then used to invert for the principal stress orientations by employing the Focal Mechanism Stress Inversion (FMSI) computer programs of Gephart. We find that the stress directions within these segments are different from each other, where they can be defined well, and the average misfits in these volumes are very small (2.8¿ and 5.5¿). The method is not robust enough to detect all the stress segmentation boundaries based on only one reference stress tensor. Analyses with different reference stress models help to define most or all the stress segmentation boundaries. ¿ American Geophysical Union 1996