The spatial distribution of hypocenters in eastern Indonesia, together with 41 new and previously published fault plane solutions, can be explained by a simple model of two lithospheric plates descending into the upper mantle beneath the Banda Sea. The major one, defined by the shallow to deep hypocenters located along the Banda arc, is a laterally continuous slab that has subducted at the plate boundary defined by the Java trench-Timor trough-Aru trough system. The other slab descends toward the southwest to depths of about 100 km in the region of the Seram trough and may be joined to the Banda arc subduction system by the westward extension of the New Guinea Tarera-Aidoena fault zone, which acts as an arc-to-arc transform. The Banda arc slab is contorted at the eastern end of the arc where the trench and the line of active volcanoes curve to the northeast. The contortion appears to be a lateral bend in the subducted slab that is continuous from the surface to depths of 600 km. Fault plane solutions in the contorted portion of the slab have P axes aligned approximately parallel to the local strike of the slab and may be related to lateral bending stresses. Similar orientations of stress axes also occur near the curved ends of other island arcs. Experiments with a lead sheet model suggest that the observed configuration of the Banda slab may be simply related to the curved end of the Banda arc and further suggest that the direction of relative motion between the Banda arc and Australia during the Neogene has been more north-northwesterly than the direction predicted from major plate motions. The discrepancy can be resolved if Southeast Asia is not considered part of the Eurasian plate. The lack of shallow underthrusting mechanisms and the paucity of shallow earthquakes in the Timor trough-Aru trough region may have resulted from the youthful collision of the Australian shelf with the Banda arc.