A systematic array processing analysis is applied to 916 teleseismic short-period P-wave signals from 25 underground explosions at Pahute Mesa in order to establish whether coherent arrivals scattered from near-source velocity heterogeneity can be identified. The signal coherence is measured using sembalance, which can be statistically evaluated using the noncentral F-distribution if the noise properties are suitable. The effects of unsuitable correlated noise due to common path and receiver contributions are suppressed by applying station weighting functions obtained by stacking all observations at a given receiver. Similar corrections for individual events are applied to reduce biases due to variations in surface reflection times and spall effects. The characteristics of the spatial and phase velocity resolution of the semblance analysis are established by a series of synthetic calculations, including cases with random arrivals in the P-wave coda. The procedure is applied to the actual data under the assumption that any scattering structures present radiate isotropically, although this constraint could be relaxed given independent knowledge of a particular scatterer's nature. The Pahute Mesa data show some evidence for enhanced scattering contributions to the first 15 sec of the P-wave signals from the western edge of the Silent Canyon Caldera. The overall confidence level in this interpretation is fairly low because simulations with random scattering in the P coda successfully match many features observed in the stacked data.