Most of our understanding of the Earth's interior has been derived from measurements from global seismic networks, although no network has ever been truly global because some 71% of the Earth's surface is underwater. The resulting gaps in coverage produce a biased and incomplete image of the Earth. Work has begun toward establishing permanent observatories on the deep seafloor, although the technical difficulties remain severe. Will these stations be useful, and where and how shall they be established? Data from seafloor observatories will be of poorer quality than continental site data because the sea surface is an important and local source of broadband noise. This noise is derived from wind and waves through direct forcing at long periods and by nonlinear coupling to elastic waves at short periods. Our understanding of the generation and propagation of seismic noise and of wind and wave climatology can be used to predict the temporal and geographical variability of the noise spectrum and to assess likely sites for permanent seafloor observatories. High noise levels near 1 Hz may raise detection limits for short-period, teleseismic arrivals above mb=7.5, limiting the usefulness of many seafloor sites. Noise levels in deep boreholes will be 10 dB quieter than those at the seafloor, but sensors buried short distances below the seafloor may also provide comparable noise levels and fidelity. The retrieval of data from permanent seafloor observatories remains an unsolved problem, but long-term temporary arrays of ocean bottom seismometers are now being used in regional scale experiments using earthquakes as sources. Such experiments are likely to be less successful in the Pacific basin than in either the Indian Ocean or North Atlantic Ocean because of higher noise levels. ¿ 1998 American Geophysical Union