Ground-motion recording obtaineds at the Van Norman Complex from the 1994 Northridge, California, mainshock and its aftershocks constitute an excellent data set for the analysis of soil response as a function of ground-motion amplitude. We searched for nonlinear response by comparing the Fourier spectral ratios of two pairs of sites for ground motions of different levels, using data from permanent strong-motion recorders and from specially deployed portable instruments. We also compared the amplitude dependence of the observed ratios with the amplitude dependence of the theoretical ratios obtained from 1-D linear and 1-D equivalent-linear transfer functions, using recently published borehole velocity profiles at the sites to provide the low-strain material properties. One pair of sites was at the Jensen Filtration Plant (JFP); the other pair was the Rinaldi Receiving Station (RIN) and the Los Angeles Dam (LAD). Most of the analysis was concentrated on the motions at the Jensen sites. Portable seismometers were installed at the JFP to see if the motions inside the structures housing the strong-motion recorders differed from nearby free-field motions. We recorded seven small earthquakes and found that the high-frequency, low-amplitude motions in the administration building were about 0.3 of those outside the building. This means that the lack of high frequencies on the strong-motion recordings in the administration building relative to the generator building is not due solely to nonlinear soil effects. After taking into account the effects of the buildings, however, analysis of the suite of strong- and weak-motion recordings indicates that nonlinearity occurred at the JFP. As predicted by equivalent-linear analysis, the largest events (the mainshock and the 20 March 1994 aftershock) show a significant deamplification of the high-frequency motion relative to the weak motions from aftershocks occurring many months after the mainshock. The weak-motion aftershocks recorded within 12 hours of the mainshock, however, show a relative deamplification similar to that in the mainshock. The soil behavior may be a consequence of a pore pressure buildup during large-amplitude motion that was not dissipated until sometime later. The motions at (RIN) and (LAD) are from free-field sites. The comparison among spectral ratios of the mainshock, weak-motion coda waves of the mainshock, and an aftershock within ten minutes of the mainshock indicate that some nonlinearity occurred, presumably at (RIN) because it is the softer site. The spectral ratio for the mainshock is between that calculated for pure linear response and that calculated from the equivalent-linear method, using commonly used modulus reduction and damping ratio curves. In contrast to the Jensen sites, the ratio of motions soon after the high-amplitude portion of the mainshock differs from the ratio of the mainshock motions, indicating the mechanical properties of the soil returned to the low-strain values as the high-amplitude motion ended. This may indicate a type of nonlinear soil response different from that affecting motion at the Jensen administration building.