Modern geodetic studies of the Yellowstone caldera, Wyoming, and its extraordinary tectonic, magmatic, and hydrothermal systems date from an initial leveling survey done throughout Yellowstone National Park in 1923 by the U.S. Coast and Geodetic Survey. A repeat park-wide survey by the U.S. Geological Survey (USGS) and the University of Utah during 1975-77 revealed that the central part of the caldera floor had risen more than 700 mm since 1923, at an average rate of 14±1 mm/yr. From 1983 to 2007, the USGS conducted 15 smaller surveys of a single level line that crosses the northeast part of the caldera, including the area where the greatest uplift had occurred from 1923 to 1975-77. The 1983 and 1984 surveys showed that uplift had continued at an average rate of 22±1 mm/yr since 1975-77, but no additional uplift occurred during 1984-85 (-2±5 mm/yr), and during 1985-95 the area subsided at an average rate of 19±1 mm/yr. The change from uplift to subsidence was accompanied by an earthquake swarm, the largest ever recorded in the Yellowstone area (as of March 2012), starting in October 1985 and located near the northwest rim of the caldera. Interferometric synthetic aperture radar (InSAR) images showed that the area of greatest subsidence migrated from the northeast part of the caldera (including the Sour Creek resurgent dome) during 1992-93 to the southwest part (including the Mallard Lake resurgent dome) during 1993-95. Thereafter, uplift resumed in the northeast part of the caldera during 1995-96, while subsidence continued in the southwest part. The onset of uplift migrated southwestward, and by mid-1997, uplift was occurring throughout the entire caldera (essentially rim to rim, including both domes). Consistent with these InSAR observations, leveling surveys indicated 24±3 mm of uplift in the northeast part of the caldera during 1995-98. The beginning of uplift was coincident with or followed shortly after an earthquake swarm near the north caldera rim during June-July 1995 - the strongest swarm since 1985. Rather than a single deformation source as inferred from leveling surveys, the InSAR images revealed two distinct sources - one beneath each resurgent dome on the caldera floor. Subsequently, repeated GPS surveys (sometimes referred to as "campaign" surveys to distinguish them from continuous GPS observations) and InSAR images revealed a third deformation source beneath the north caldera rim. The north-rim source started to inflate in or about 1995, resulting in as much as 80 mm of surface uplift by 2000. Meanwhile, motion of the caldera floor changed from uplift to subsidence during 1997-8. The north rim area rose, while the entire caldera floor (including both domes) subsided until 2002, when both motions paused. Uplift in the northeast part of the caldera resumed in mid-2004 at a historically unprecedented rate of as much as 70 mm/yr, while the north rim area subsided at a lesser rate. Resurveys of the level line across the northeast part of the caldera in 2005 and 2007 indicated the greatest average uplift rate since the initial survey in 1923-53±3 mm/yr. Data from a nearby continuous GPS (CGPS) station showed that the uplift rate slowed to 40-50 mm/yr during 2007-8 and to near zero by September 2009. Following an intense earthquake swarm during January-February 2010, this one near the northwest caldera rim and the strongest since the 1985 swarm in the same general area, CGPS stations recorded the onset of subsidence throughout the entire caldera. Any viable model for the cause(s) of ground deformation at Yellowstone should account for (1) three distinct deformation sources and their association with both resurgent domes and the north caldera rim; (2) interplay among these sources, as suggested by the timing of major changes in deformation mode; (3) migration of the area of greatest subsidence or uplift from the northeast part of the caldera to the southwest part during 1992-95 and 1995-97, respectively; (4) repeated cycles of uplift and subsidence and sudden changes from uplift to subsidence or vice versa; (5) spatial and temporal relationships between changes in deformation mode and strong earthquake swarms; and (6) lateral dimensions of all three deforming areas that indicate source depths in the range of 5 to 15 km. We prefer a conceptual model in which surface displacements at Yellowstone are caused primarily by variations in the flux of basaltic magma into the crust beneath the caldera. Specifically, we envision a magmatic conduit system beneath the northeast part of the caldera that supplies basalt from a mantle source to an accumulation zone at 5-10 km depth, perhaps at a rheological boundary within a crystallizing rhyolite body remnant from past eruptions. Increases in the magma flux favor uplift of the caldera and decreases favor subsidence. A delicate equilibrium exists among the mass and heat flux from basaltic intrusions, heat and volatile loss from the crystallizing rhyolite body, and the overlying hydrothermal system. In the absence of basalt input, steady subsidence occurs mainly as a result of fluid loss from crystallizing rhyolite. At times when a self-sealing zone in the deep hydrothermal system prevents the escape of magmatic fluid, the resulting pressure increase contributes to surface uplift within the caldera; such episodes end when the seal ruptures during an earthquake swarm. To account for the north rim deformation source, we propose that magma or fluid exsolved from magma episodically escapes the caldera system at the three-way structural intersection of (1) the northern caldera boundary, (2) an active seismic belt to the north-northwest that is associated with the Hebgen Lake fault zone, and (3) the Norris - Mammoth corridor - a zone of faults, volcanic vents, and thermal activity that strikes north from the north rim of the caldera near Norris Geyser Basin to Mammoth Hot Springs near the northern boundary of Yellowstone National Park. Increased fluid flux out of the caldera by way of this intersection favors subsidence of the north rim area, and decreased flux favors uplift. This model does not account for poroelastic and thermoelastic effects, nonelastic rheology, or heat and mass transport in the hot and wet subcaldera crust. Such effects almost surely play a role in caldera deformation and are an important topic of ongoing research.