We used vertical seismic profiling (vsp) data collected in shallow boreholes (about 40–60 m deep) to determine the shear-wave velocity at four sites in the Mississippi embayment in southwestern Tennessee. The source was an air-powered hammer that produces repeatable SH waves, which were recorded by source monitor geophones deployed on the surface very close to the source. Three approaches were used to determine interval velocities: an approximate zero-offset method, a layer- stripping method, and a waveform-matching method. The first two methods use arrival-time picks, whereas the latter is based on the fit of synthetic vsp data to the first half-cycle (approximately) of each trace. The advantage of this method over the other two is that it uses a segment of the data, rather than a single data point. Therefore, the velocities determined using the waveform-matching method are better constrained and are not affected by picking errors, which may translate into significant spurious velocity variations. The source wavelets recorded by one of the monitor geophones and the velocity model computed with the layer-stripping method were used to generate synthetic vertical seismic profiling data for comparison with the actual data. Then the model velocities were modified interactively, one layer at a time, until a satisfactory match was achieved. This required including attenuation in the computation of the synthetic data. The four sites investigated in this study can be divided into two groups: low-velocity sites (Shelby Farms and Covington) and high-velocity sites (Brownsville and Jackson). These last two sites are at larger distances from the embayment axis than the other two, which means that the difference in velocities probably corresponds to the presence of different geologic units. Good agreements between the lithology in the boreholes and the velocity profiles were obtained for all the four sites.