We compare microtremor data to weak-motion S-wave and coda recordings at sites in San Francisco in order to clarify the range of applicability of microtremor data to ground-motion prediction. We also compare S-wave results to coda results. For each type of data, we compute spectral ratios of motions from two soil/rock station pairs and from an uphole/downhole pair in the Marina district. We compute horizontal/vertical ratios (Nakamura's method) at a soil site, a rock site, and the surface and borehole instruments. In the station-pair analyses, microtremor data show amplifications at the same fundamental frequency as S waves, but the frequencies of other peaks do not agree. The amplification at frequencies higher than 2 Hz is greater in the microtremor data. Station-pair ratios of coda data generally show spectral peaks occurring at the same frequencies, but with levels varying from one to four times the amplification from S-wave ratios. Nakamura's method of analyzing microtremors agrees better with S-wave station-pair results than the microtremor station-pair method over a limited frequency band that varies from station to station.