In 1991, during an experiment to compare low-frequency seismic noise on a basaltic and a sediment covered seafloor (NOBS), we recorded teleseisms on the Juan de Fuca ridge, the Gorda ridge, and the adjacent Cascadia Basin with the SNAG ocean-bottom seismometers (OBS). These data provide an indication of the type of data that may be obtained from future experiments to record teleseisms and may be helpful in designing these experiments and analyzing the results. We found that although seafloor noise is dominated by microseisms in the band 0.1 to 0.3 Hz, there is a well-developed minimum in noise from about 0.03 to 0.1 Hz (the noise notch). In this noise notch, teleseisms can be most easily detected. In the Cascadia area, the overall noise levels are such that only teleseismic events with magnitude greater than 6.5 were usefully recorded. A magnitude 6.6 event in the New Britain area(Delta = 89 degrees) produced usable P- and surface-wave data only in this noise notch. In the band 0.03 to 0.1 Hz, the character of compressional waves is very sensitive to water depth and the type of sensor. We show that pressure sensors are especially sensitive to reverberation in the ocean and that motion sensors (seismometers) are less sensitive to ocean reverberations and will record teleseismic phases with less distortion than pressure sensors. The Cascadia data indicate enhanced P amplitudes at sites on the ridge axes that could be due to focusing caused by a low-velocity lens. These data suggest that amplitude information may be as, or even more, useful than P delay times for determining upper mantle structure.