We develop an earthquake travel-time inversion methodology suitable for determining three-dimensional velocity structure and fault-plane orientation for an area with limited a priori information. Using a cascaded combination of a nonlinear simulated annealing optimization and linearized inversion, we investigate local three-dimensional compressional velocity structure and estimate the orientation of a fault plane in the Eureka Valley area of eastern California. We inverted travel times recorded at 20 permanent and 8 portable stations from an M 6.1 mainshock and a few hundred aftershocks for P-wave velocity and hypocentral coordinates. Using the velocity model obtained by the nonlinear optimization as an initial model for linearized inversion, we relocated the hypocenters and further fine-tuned the model. The relocated hypocenters define a north-northwest-trending fault dipping steeply westward. The final crustal velocity model features a low-velocity trend along the strike of the Eureka Valley and a high-velocity block southwest of the valley. Compared with a fully linearized inversion, our scheme demonstrates independence of the final results on the initial model and the potential of avoiding local minima.