We model the source of inflation of Long Valley caldera by combining geodetic and micro-gravity data. Uplift from GPS and leveling, two-color EDM measurements, and residual gravity change determinations are used to estimate the intrusion geometry, assuming a vertical prolate ellipsoidal source. The U.S. Geological Survey occupied the Long Valley gravity network six times from 1980 to 1985. We reoccupied this network twice, in the summer of 1998 (33 stations), and the summer of 1999 (37 stations). Before gravity data can be used to estimate the density of the intrusion, they must be corrected for the effect of vertical deformation (the free-air effect) and changes in the water table. We use geostatistical techniques to interpolate uplift and water table changes at the gravity stations. The inflation source (a vertical prolate ellipsoid) is located 5.9 km beneath the resurgent dome with an aspect ratio equal to 0.475, a volume change from 1982 to 1999 of 0.136 km3 and a density of around 1700 kg/m3. A bootstrap method was employed to estimate 95% confidence bounds for the parameters of the inflation model. We obtained a range of 0.105-0.187 km3 for the volume change, and 1180-2330 kg/m3 for the density. Our results do not support hydrothermal fluid intrusion as the primary cause of unrest, and confirm the intrusion of silicic magma beneath Long Valley caldera. Failure to account for the ellipsoidal nature of the source biases the estimated source depth by 2.9 km (a 33% increase), the volume change by 0.019 km3 (a 14% increase) and the density by about 1200 kg/m3 (a 40% increase). ?? 2003 Elsevier B.V. All rights reserved.