Since 1986, nine years of wave data derived from satellites have been accumulated, and this database will expand dramatically in the next two years as two more satellites are added. Several researchers have begun using this data to estimate extreme value statistics for waves. However, one potential problem with satellite data is space-time resolution, which is a poor match for the scales of storms. Satellites only revisit a site once every 10-35 days, and their tracks are separated by 100-200 km. With this coarse sampling, the satellite may miss storms since they have characteristic length and time scales as short as a few hours and tens of kilometers. The purpose of this paper is to explore the impact of this undersampling on the calculated 100-yr wave height. This is accomplished by running Monte Carlo simulations of simplified but realistic storms sampled by a simulated satellite and site. The authors study the sensitivity of the calculated 100-yr wave to variations in storm type, radius, and forward speed; number of satellites; satellite track; and satellite sampling region. The uncertainty, as measured by the coefficient of variation (cov), of the 100-yr wave based on 10 years of satellite data is 10% in regions like the North Sea that are dominated by extratropical storms, provided the satellite data is sampled over a 200-300-km region. This is about the level accepted by present offshore standards like the American Petroleum Institute. For regions dominated by tropical storms like the Gulf of Mexico, the cov for satellite- or site-derived extremes is much greater than 10% using 10 years of data. The situation improves with increased sample period, storm frequency, or the number of satellites. However, even in these cases some caution must still be exercised near the coast where the satellite data itself may be less reliable and sampling over large regions may remove real spatial gradients. Our conclusions apply to all existing satellite tracks including Geosat, Topex/Poseidon, and ERS.