The classic double seismic zone in the central portion of the Kuril‐Kamchatka arc seems to turn into a regular Wadati‐Benioff zone with downdip compression and extension in the northeastern and southwestern ends of the arc, respectively. To delineate this lateral changeover, we studied 33 large‐ to moderate‐sized (mb≥5.1) earthquakes that occurred at depths shallower than 200 km between 1963 and 1991. These events took place in two regions where the double seismic zone terminates and single seismic zones emerge. Precise earthquake source parameters, including focal depths and mechanisms, are determined from a careful analysis of a mixture of analog and digital P and SH waveforms recorded at teleseismic distances. Our results indicate that the double seismic zone is confined between approximately 46° and 53°N. Northeast of 53°N, intermediate‐depth earthquakes (∼70–200 km) show only downdip compressional focal mechanisms. Furthermore, the Wadati‐Benioff zone there seems to be a continuation of the upper, compressional layer of the double seismic zone. Thus the changeover from a double to a single seismic zone in this case is simply disappearing of the lower, extensional sheet of the double seismic zone. Similarly, southwest of 46°N, intermediate‐depth earthquakes show only downdip extensional mechanisms. The single seismic zone is unlikely to consist of two layers of extensional seismicity. In fact, nowhere along the entire arc did we observe two distinct seismogenic layers with a single state of strain. In other words, these extensional events did not occur in the same seismogenic structure as the upper (compressional) layer of the double seismic zone. However, we cannot ascertain whether the single seismic zone in extension is simply a southwestern extension of the lower (extensional) layer of the double seismic zone. Lateral variation from a single zone under extension in the southwest, to a double seismic zone, and finally to a single zone under compression cannot be accounted for by conventional models of a double seismic zone. On the other hand, the addition of a component of stress transmitted from depths greater than 250 km can explain the observed lateral variation in the state of strain at intermediate depths because the slab between depths of 250–450 km is under extension in the southwestern end of the arc, whereas the rest of the arc is under compression. Vanishing of the lower (extensional) layer of the double seismic zone toward the northeastern end of the arc is interpreted as a result of compression transmitted from below overshadowing extension caused by effects such as unbending. A similar reasoning applies to the southwestern termination of the double seismic zone.