To investigate the morphology of subducted slab in the mantle below northwest Pacific island arcs we inverted traveltime residuals for aspherical variations in P-wave propagation velocity relative to the radially symmetric iasp91 reference model. The tomographic method used is based on a step-wise linearization of the inversion problem. First, we relocated ISC (International Seismological Centre) hypocentres with re-identified P and pP phase data using the iasp91 traveltime tables. The variance of P residuals relative to iasp91 traveltimes was 17 per cent less than the variance of P data reported by the ISC relative to the Jeffreys-Bullen (J-B) traveltime tables. Second, we performed a linearized (LSQR) inversion for Earth structure and source relocation with the P and pP residuals obtained from the first step, using iasp91 as the reference model for seismic velocities. The incorporation of the depth phase pP in the tomographic inversions has two major advantages: (1) the pP data provide constraints on focal depth and thus reduce the trade-off between source relocation and structure; and (2) the pP ray paths improve the sampling of Earth structure in the shallow mantle and transition zone. We used more than 2 times 10⁶ and about 1 times 10⁵ P- and pP-wave traveltime residuals, respectively, from about 40 000 earthquakes with epicentres in the study region that were recorded at one or more of the 2300 globally distributed seismological stations considered in this study. We assessed the spatial resolution in the tomographic images with checker board-type sensitivity tests. These tests reveal high resolution of upper mantle and transition-zone structure, particularly below the central part of our study region. Structure with wavelengths of the order of 100 km is resolved below Japan, whereas structure with wavelengths of the order of 300 km is well resolved below the Kuril, Izu Bonin and Ryukyu arcs. Small-scale structure is poorly resolved in depth below the northern part of the Kuril-Kamchatka arc and below the Izu Bonin and Mariana arcs. This limits the interpretation of slab structure and mantle flow from tomographic images alone. With this limitation in mind, we conclude from the tomographic images that subducted slab deflects in the mantle transition zone below the geographical area encompassed by the Kuril basin, the Japan Sea, and the northern part of the Philippine Sea. This is in good agreement with the results of other recently published tomographic studies, the occurrence of earthquakes several hundred kilometres off the inclined Wadati-Benioff seismic zones, and inferences about ‘660 km’ discontinuity topography. In contrast, slab-like structures of high P-wave velocity are imaged in the lower mantle below the deepest earthquakes of the northern Kuril-Kamchatka and Mariana seismic zones. This is indicative of local slab penetration of the lower mantle. From tomographic images we cannot discern between compositionally or thermally induced variations in seismic velocity. However, with regard to the nature of the boundary between upper and lower mantle, our observations argue against either compositional mantle layering with large contrasts in intrinsic density or phase changes with steep Clapeyron slopes.