Since Bezymianny Volcano resumed its activity in 1956, eruptions have been frequent; recently with up to 1–2 explosive events per year. To investigate deformation related to this activity we installed a GPS network of 8 continuous and 6 campaign stations around Bezymianny. The two striking observations for 2005–2010 are (1) rapid and continuous network-wide subsidence between 8 and 12 mm/yr, which appears to affect KAMNET stations more than 40 km away where we observe 4–5 mm/yr of subsidence, and (2) only the summit station BZ09 shows slight deviations from the average motion in the north component at times of eruptions. The network-wide subsidence cannot be explained by tectonic deformation related to the build-up of interseismic strain due to subduction of the Pacific plate. A first order model of surface loading by eruptive products of the Kluchevskoy Group of Volcanoes also explains only a fraction of the subsidence. However, a deep sill at about 30 km under Kluchevskoy that constantly discharges material fits our observations well. The sill is constrained by deep seismicity which suggests 9.5 km width, 12.7 km length, and a 13° dip-angle to the south-east. We infer a closing rate of 0.22 m/yr, which implies a volume loss of 0.027 km3/yr (0.16 m/yr and 0.019 km3/yr considering surface loading). Additional stations in the near and far field are required to uniquely resolve the spatial extent and likely partitioning of this source. We explain the eruption related deformation at BZ09 with a very shallow reservoir, likely within Bezymianny's edifice at a depth between 0.25 km and 1.5 km with a volume change of 1–4 × 10− 4 km3. Much of the material erupted at Bezymianny may be sourced from deeper mid-crustal reservoirs with co-eruptive volume changes at or below the detection limit of the GPS network. Installation of more sensitive instruments such as tiltmeters would allow resolving of subtle co-eruptive motion.