In spite of an order of magnitude increase over the past 15 years in spatial sampling of the wavefield, a major uncertainty in the analysis of active source seismic data remains phase identification. This uncertainty results in part from the wide range of spatial scales of velocity heterogeneity in the crust. Smaller scale variations than those which can be deterministically resolved given the design of a particular seismic experiment can be modeled statistically using geologic constraints. Here we present synthetic seismograms generated from several different realizations of a stochastic model describing the velocity heterogeneity of Franciscan terrane rocks. We compare the results to observed data and to synthetic seismograms generated for a model derived from tomographic inversion of the data in order to obtain qualitative insights into the relative importance of large and small scale velocity heterogeneity. Not surprisingly, the synthetic data for the tomographic model best reproduce observed small-scale variations in first arrival time, which only occur for particular realizations of the stochastic model. The synthetic seismograms generated for the stochastic models best reproduce the level of signal-generated noise and suggest that the amplitude of velocity variation locally within the Franciscan is approximately 1 km/s. They also illustrate the effect of a strongly heterogeneous upper and mid-crust on the amplitude-versus-offset pattern of arrivals from the lower crust and upper mantle. These effects may sometimes be interpreted deterministically, leading to biased models or an overly optimistic estimate of lower crustal resolution. ¿ 1997 American Geophysical Union