Empirical Green's functions (EGFs) between pairs of seismographs can be estimated from the time derivative of the long-time cross-correlation of ambient seismic noise. These EGFs reveal velocity dispersion at relatively short periods, which can be used to resolve structures in the crust and uppermost mantle better than with traditional surface-wave tomography. We combine Rayleigh-wave dispersion estimates from EGFs and from traditional two-station (TS) analysis into a new approach to surface-wave array tomography with data from dense receiver arrays. We illustrate the methodology with continuous broad-band recordings from a temporary seismographic network on the southeastern part of the Tibetan plateau, in Sichuan and Yunnan provinces, SW China. The EGFs are robust under temporal changes in regional seismicity and the use of either ambient noise (approximated by records without signal from events with magnitude mb≥ 5 or 4) or surface wave coda produces similar results. The EGFs do not strongly depend on the presence of large earthquakes, but they are not reciprocal for stations aligned in the N–S direction. This directionality reflects the paucity of seismicity to the north of the array. Using a far-field representation of the surface-wave Green's function and an image transformation technique, we infer from the EGFs the Rayleigh-wave phase velocity dispersion in the period band from 10–30 s. A classical TS approach is used to determine Rayleigh-wave phase velocity dispersion between 20–120 s. Together, they constrain phase velocity variations for T= 10–120 s, which can be used to study the structure from the crust to the upper mantle. Beneath SE Tibet, short and intermediate period (10–80 s) phase velocities are prominently low, suggesting that the crust and upper mantle beneath SE Tibet is characterized by slow shear wave propagation.