This paper represents a search for a set of conditions necessary and sufficient to enhance seismicity upon reservoir filling. In the small number of reservoirs that have clearly been seismogenic, criticality may have been attained beforehand by recent stress changes. These can be deduced from the erosion of landforms, the deposition of sediments, crustal uplift and geostatic cooling or subsidence and heating, as well as from pore-pressure changes resulting from river incision and alluviation. Geomorphic evolution has produced effective-stress changes greater in magnitude than the superposed stresses from reservoir filling. Increased total stresses due to reservoir weight extend to lesser distances than the increased fluid pressures imposed by changes of the regional ground-water flow system. Thus the field of reduced effective stress, wherein failure by fault slippage can occur, depends upon the hydrogeologic setting, which is systematically examined in this report. Widely spaced stiff fractures in unweathered basement rock are essential to reduction of effective stresses. Consolidation shortens the fluid-transient period. Field evidence suggests that Poisson’s ratio for extensive rock masses exceeds 0.33, a parameter needed for estimating horizontal stress changes from vertical changes. Application of theoretical stress changes in different tectonic environments suggests that natural and man-made seismicity may be influenced by surficial loading. Because the cumulative changes of principal stresses are the result of compensative stabilizing and destabilizing effects of total load and fluid pressure, each case needs to be analyzed individually. Even in thrust-fault regions, reservoir filling might stimulate seismicity. Kariba Reservoir in Africa exemplifies induced seismicity in a normal-fault environment. Faults in the Karoo-filled, ancient mid-Zambezi rift have been inactive since Cretaceous time, but minor pre-reservoir seismicity attests to criticality, which was maintained by erosion. The river lowered the valley floor without faulting, whereas in other rifts, topographic similarity and basement stresses were maintained by fault displacements. The existence of faults, artesian aquifers, and thermal springs submerged by Kariba Reservoir imply that upon filling, fluid pressures were increased over distances as much as several reservoir widths away from the site and to depths of several kilometres. Whereas post-reservoir seismicity was produced by decrease of effective stress, subsidence was responsive to increase of total stress. Koyna Reservoir in India exemplifies induced seismicity in a wrench-fault environment. The site is topographically uplifted and close to the western fault-boundary of the Indian peninsula. The flat-lying Deccan lava flow contacts, together with vertical north-south wrench faults, promote extensive ground-water circulation from the reservoir to the coastal lowland. Uplift and erosion maintained criticality and minor prior seismicity. Reservoir filling renewed faulting by increasing the pore pressure in the system. A major earthquake swarm, culminating in a disastrous M = 6.7 event, propagated down the hydraulic gradient. Progressive decrease of effective stress is inadequate as a criterion for seismogenicity of reservoirs because it is also satisfied by many nonseismic reservoirs. Strength and pore-pressure measurements deep beneath reservoir shores are not recommended; the former are too imprecise, the latter too small for utility. Seismic prediction may depend in the future on techniques of structural discrimination. Heterogeneity of the crust is the main problem: large areas are stressed subcritically, probably because the weakest faults limit the stresses transmitted. Among the factors needing study are differences in fault materials, contact geometry and continuity, and fault orientation with respect to stresses. If analysis of historic and predicted stress changes satisfy necessary but not sufficient criteria for failure, then until means are available to identify criticality, the information can find practical application in adjustment of probabilistic measures of historic seismicity.