We deployed a dense seismic array to image the shallow structure in the injection area of the Brady Hot Springs geothermal site in Nevada. The array was composed of 238 5 Hz, three-component nodal instruments and 8,700 m of distributed acoustic sensing (DAS) fiber-optic cable installed in surface trenches plus 400 m installed in a borehole. The geophone array had about 60 m instrument spacing in the target zone, whereas DAS channel separations were about 1 m. The acquisition systems provided 15 days of continuous records including active source and ambient noise signals. A large vibroseis truck (T-Rex) was operated at 196 locations exciting a swept-frequency signal from 5 to 80 Hz over 20 seconds using three vibration modes. Sweeps were repeated up to four times during different modes of geothermal plant operation: normal operation, shutdown, high and oscillatory injection and production, and normal operation again. The cross- correlation method was utilized to remove the sweep signal from the geophone records. The first P arrivals were automatically picked from the cross-correlation functions using a combination of methods, and the travel times were used to invert for the 3D P-wave velocity structure. Models with 50 m horizontal node spacing were obtained, with vertical node spacing of 10 to 50 m. The travel time data were fit to about 30 ms, close to our estimated picking uncertainty. Boundaries between high and low velocity zones agree with previous surveys of local faults and low velocity zones near the surface correspond to fumarole locations. A rapid increase in velocity at about 50 m depth fits with borehole data on the depth of the Quaternary sediments. There is some evidence for changes in the P-wave velocity during the experiment with slower travel times at the beginning of the experiment.