We obtain a three-dimensional (3D) model of shear wave velocity heterogeneity of the Earth's mantle by inverting a large set of seismic data consisting of 27,000 long-period seismograms and 14,000 travel time observations. About 60% of the data has been collected through the efforts of several research groups and used in earlier studies. The new data, which come from stations of different seismic networks including the Chinese Digital Seismographic Network (CDSN) and Geoscope, are extracted to provide sampling of mantle heterogeneity as uniform as possible. Because of the improved data coverage, we expand our model to degree 12 in spherical harmonics to describe horizontal variations, and to order of 13 in Chebyshev polynomials to describe radial variations. The resulting model shows a clear pattern of slower-than-average shear velocities at shallow depths underlying the major segments of the world-wide ridge system. These anomalies extend to depths greater than 300 km and in some cases appear to continue into the lower mantle. There is also a good correlation between the major continental shields and fast-velocity perturbations at depths extending to 300--400 km. Some of the continental ''roots'' extend to depths greater than in other studies. The pattern of heterogeneity is more complex in the midmantle, where the power spectrum is almost flat and has a relatively low amplitude; therefore the results in this depth range should be interpreted with caution.

The pattern of the heterogeneity indicates a rapid change at a depth of about 1700 km. At this depth, the power spectrum of the model shifts from one which is almost flat in the midmantle to that dominated by degrees 2 and 3; this pattern then continues to the core-mantle boundary (CMB). The model is dominated by a few megastructures of velocity heterogeneity below the depth of 2000 km, in agreement with previous studies. Among these megastructures are the ''Pangea Trough,'' ''Great African Plume,'' and ''Equatorial Pacific Plume Group.'' The model predicts well the large-scale pattern of observed S, SS absolute travel times, and SS-S, ScS-S differential travel times. It also predicts well the waveforms of mantle wave and body wave. We compare our model with several other recently published models. There is generally a good agreement in the long-wavelength pattern of the models, especially at shallow depths and near the CMB. However, the amplitude as well as the pattern of shorter-wavelength features are in some cases quite different.