Abstract The mantle electromagnetic and tomography (MELT) experiment on the east Pacific rise near 17°S was the first large teleseismic experiment on a midocean ridge. During the six-month deployment, no compressional arrivals were well recorded above 0.5 Hz. In comparison, the ICEMELT experiment in Iceland recorded compressional arrivals at 1-2 Hz from about 2 earthquakes per month. We compare noise spectra from the two experiments and show that this difference in detection is at least in part a result of noise. Near 1 Hz, seismic noise in the oceans is produced locally by wind-generated waves. At both experiment sites, 1-Hz noise levels are well correlated with local sea-surface-wind speeds derived from satellite observations. For a given wind speed, 1-Hz noise levels are about 10-20 dB lower in Iceland. At the MELT site, cross-correlations of wind speed with the logarithm of noise in a narrow-frequency band yield correlation coefficients exceeding 0.7 at frequencies between 0.4 Hz and 2 Hz. Noise levels at 1 Hz increase with wind by 1.3-1.4 dB per m/sec for wind speeds less than 10 m/sec. For the ICEMELT experiment, high correlation coefficients extend to markedly higher frequencies for coastal stations, and there is a 10-dB drop in 1-Hz noise levels 100-km inland. Noise levels increase by about 0.8 dB per m/sec. The strong correlation between wind speed and 1-Hz seismic noise provides justification for using satellite wind speed data to search for locations on the global spreading system where there is a better probability of recording high-frequency arrivals. The calmest sites are found on the northern east Pacific rise, near the equator in all oceans, and near 34° N and 22° S on the mid-Atlantic ridge.