Recent global travel time tomography studies by Zhou <1996> and van der Hilst et al. <1997> have been performed with cell parameterizations of the order of those frequently used in regional tomography studies (i.e., with cell sizes of 1¿--2¿). These new global models constitute a considerable improvement over previous results that were obtained with rather coarse parameterizations (5¿ cells). The inferred structures are, however, of larger scale than is usually obtained in regional models, and it is not clear where and if individual cells are actually resolved. This study aims at resolving lateral heterogeneity on scales as small as 0.6¿ in the upper mantle and 1.2¿--3¿ in the lower mantle. This allows for the adequate mapping of expected small-scale structures induced by, for example, lithosphere subduction, deep mantle upwellings, and mid-ocean ridges. There are three major contributions that allow for this advancement. First, we employ an irregular grid of nonoverlapping cells adapted to the heterogeneous sampling of the Earth's mantle by seismic waves <Spakman and Bijwaard, 1998>. Second, we exploit the global data set of Engdahl et al. <1998>, which is a reprocessed version of the global data set of the International Seismological Centre. Their reprocessing included hypocenter redetermination and phase reidentification. Finally, we combine all data used (P, pP, and pwP phases) into nearly 5 million ray bundles with a limited spatial extent such that averaging over large mantle volumes is prevented while the signal-to-noise ratio is improved. In the approximate solution of the huge inverse problem we obtain a variance reduction of 57.1%. Synthetic sensitivity tests indicate horizontal resolution on the scale of the smallest cells (0.6¿ or 1.2¿) in the shallow parts of subduction zones decreasing to approximately 2¿--3¿ resolution in well-sampled regions in the lower mantle. Vertical resolution can be worse (up to several hundreds of kilometers) in subduction zones with rays predominantly pointing along dip. Important features of the solution are as follows: 100--200 km thick high-velocity slabs beneath all major subduction zones, sometimes flattening in the transition zone and sometimes directly penetrating into the lower mantle; large high-velocity anomalies in the lower mantle that have been attributed to subduction of the Tethys ocean and the Farallon plate; and low-velocity anomalies continuing across the 660 km discontinuity to hotspots at the surface under Iceland, east Africa, the Canary Islands, Yellowstone, and the Society Islands. Our findings corroborate that the 660 km boundary may resist but not prevent (present day) large-scale mass transfer from upper to lower mantleor vice versa. This observation confirms the results of previous, global mantle studies that employed coarser parameterizations. ¿ 1998 American Geophysical Union