Large data sets of vertical and horizontal seismograms from the Pacific Northwest Seismic Network, Northern California Seismic Network, and Berkeley Digital Seismic Network are used to study the high-frequency (0.25–16 Hz) ground-motion scaling characteristics in Washington–Oregon, northern California (39° N to 42° N and 119° W to 124° W), and central California (35° N to 39° N and 118° W to 123° W). We used peak filtered ground velocities to characterize the propagation, excitation, and site terms. The regression results for propagation were modeled using a geometrical spreading function, g(r), and a frequency-dependent attenuation, Q(f)=Q0fη. For the Pacific Northwest, the best Q model that fits the observation is expressed by Q(f)=280f0.55. The eastern central California and western central California results are parameterized with Q(f)=280f0.50 and Q(f)=240f0.35, respectively. The northern California results are not easy to model and require using a frequency-dependent η and a frequency-dependent geometrical spreading. The geometrical spreading effect for the frequencies higher than 5 Hz is very strong in that region. The excitation terms for the small events studied were modeled using a Brune’s source model. An average stress drop of 30 bars was obtained for the Pacific Northwest. The northern California average value for stress drop is 90 bars. The observations of eastern central California were modeled with a stress drop of 49 bars, while the best fit for the western part required Δσ=80 bars. The range of the values obtained for Q, g(r), and Δσ indicates that the ground-motion parameters for one region should not be used for another. As an example, our results show that the ground-motion amplitude due to an Mw 5.0 earthquake at a distance of 50 km is different by a factor of 1.5 between the Pacific Northwest and eastern central California.