The observed power spectrum of the earth's gravity field is interpreted as resulting from randomly distributed density anomalies of relatively small spatial coherence occurring throughout at least the first 800-1000 km of the mantle. The magnitude of these density anomalies is of the order of 1.7¿105/(dL dr)1/2 g cm-3. where dL (in centimeters) is their horizontal spatial coherence and dr the vertical coherence. Most of the power in harmonics of degree >6 reflects heteorogeneities in the first 300-400 km of the mantle. The lower-degree harmonics reflect conditions down to about 800-1000 km. Anomalies in density of the above magnitude occurring slightly to the nonhydrostatic potential of degrees 2 and 3. This anomalous density estimate therefore presents an upper limit to permissable values that can occur in the lower mantle. When Bich's law is used this indicates that lateral anomalies in the velocity of P waves are of the order of 0.2% (rms value) or less, in general smaller than the seismic results for the lower mantle. The gravity signal of these lower mantle anomalies is masked by that of upper mantle anomalies, a fact which explains the absence of correlation found between gravity and deep mantle seismic anomalies. The seismic anomalies reported for the upper mantle often exceed the rms value of 0.1-0.3% estimated from the gravity field. This indicates that internal compensation occurs whereby mass excesses are underlain or overlain by mass deficits. Other estimates deduced from the gravity field are upper limits to the lateral temperature variations of 100¿C (rms value) and upper limits to permissible topography on the core-mantle boundary of the order of 200 m or less. Geopotential harmonics of degree >6 correlate significantly with the surface topography although linear regression analysis shows that only about 10% of the total power of the observed geopotential is responsible for this correlation. This is of the same magnitude as the power in the potential spectrum computed from the topography and its isostatic compensation and suggests that this correlation is a consequence of near-surface effects rather than deeper mantle effects.