The effect of alluvium on strong ground motion can be seen by comparing two strong-motion records of the Coyote Lake, California, earthquake of 6 August 1979 (ML = 5.9). One record at a site on Franciscan bedrock had a peak horizontal acceleration of 0.13 g and a peak horizontal velocity of 10 cm/sec. The other, at a site 2 km distant on 180 meters of Quaternary alluvium overlying Franciscan, had values of 0.26 g and 32 cm/sec, amplifications by factors of 2 and 3. Horizontal motions computed at the alluvial site for a linear plane-layered model based on measured P and S velocities show reasonably good agreement in shape with the observed motions, but the observed peak amplitudes are greater by a factor of about 1.25 in acceleration and 1.8 in velocity. About 15 per cent of the discrepancy in acceleration and 20 per cent in velocity can be attributed to the difference in source distance; the remainder may represent focusing by refraction at a bedrock surface concave upward. There is no clear evidence of nonlinear soil response. Fourier spectral ratios between motions observed on bedrock and alluvium show good agreement with ratios predicted from the linear model. In particular, the observed frequency of the fundamental peak in the amplification spectrum agrees with the computed value, indicating that no significant nonlinearity occurs in the secant shear modulus. Computations show that nonlinear models are compatible with the data if values of the coefficient of dynamic shear strength in terms of vertical effective stress are in the range of 0.5 to 1.0 or greater. The data illustrate that site amplification may be less a matter of resonance involving reinforcing multiple reflections, and more the simple effect of the low near-surface velocity. Application of traditional seismological theory leads to the conclusion that the site amplification for peak horizontal velocity is approximately proportional to the reciprocal of the square root of the product of density and shear-wave velocity.