Ground motion from local earthquakes and the SHIPS (Seismic Hazards Investigation in Puget Sound) experiment is used to estimate site amplification factors in Seattle. Earthquake and SHIPS records are analyzed by two methods: (1) spectral ratios relative to a nearby site on Tertiary sandstone, and (2) a source/site spectral inversion technique. Our results show site amplifications between 3 and 4 below 5 Hz for West Seattle relative to Tertiary rock. These values are approximately 30% lower than amplification in the Duwamish Valley on artificial fill, but significantly higher than the calculated range of 2 to 2.5 below 5 Hz for the till-covered hills east of downtown Seattle. Although spectral amplitudes are only 30% higher in the Duwamish Valley compared to West Seattle, the duration of long-period ground motion is significantly greater on the artificial fill sites. Using a three-dimensional displacement response spectrum measure that includes the effects of ground-motion duration, values in the Duwamish Valley are 2 to 3 times greater than West Seattle. These calculations and estimates of site response as a function of receiver azimuth point out the importance of trapped surface-wave energy within the shallow, low-velocity, sedimentary layers of the Duwamish Valley. One-dimensional velocity models yield spectral amplification factors close to the observations for till sites east of downtown Seattle and the Duwamish Valley, but underpredict amplifications by a factor of 2 in West Seattle. A two-dimensional finite-difference model does equally well for the till sites and the Duwamish Valley and also yields duration estimates consistent with the observations for the Duwamish Valley. The two-dimensional model, however, still underpredicts amplification in West Seattle by up to a factor of 2. This discrepancy is attributed to 3D effects, including basin-edge-induced surface waves and basin-geometry-focusing effects, caused by the proximity of the Seattle thrust fault and the sediment-filled Seattle basin.