It was late on the morning of April 1, 1946, and on the island of Hawaii, children from the school at Laupahoehoe Point were the first to see the Pacific Ocean disappear. They watched, awestruck, as 500 feet of sand and coral emerged glistening into the sunshine. A few of the braver ones ventured out onto the exposed reef. Suddenly the water came roaring back, sweeping away the children along with the buildings near the shore and the entire waterfront of nearby Hilo. For nine hours, a teacher, 21-year-old Marsue McGinnis, clung to a piece of driftwood before she was spotted by her fiance, who had mounted his own rescue in a borrowed motorboat. "I saw a number of children floating near me, clinging to wreckage," she said. "We just kept floating out to sea, and some of the children disappeared."

Five hours earlier, an earthquake had erupted under the ocean floor off the coast of Alaska. Officials of the Coast and Geodetic Survey, a branch of the Commerce Department, knew what might be coming toward Hawaii, even if they didn't know enough to call it a tsunami, the technical term from the Japanese that only came into widespread use in America in the 1960s. But, as they blandly informed reporters the next day, no warning was issued because it was "impossible to predict whether the waves would be large enough to cause damage." Since then, great strides have been made in both geophysics and the concept of government accountability. The Pacific Tsunami Warning Center, established in Hawaii two years later, has issued warnings for all five of the significant ocean-spanning tsunamis over the past 56 years--and for 15 others that turned out to be false alarms. It correctly predicted that the magnitude 9.5 earthquake off the coast of Chile on May 22, 1960--the most powerful of the 20th century--would generate dangerous tsunamis in Hawaii and as far away as Japan. Unfortunately, human nature is more resistant to change than either technology or bureaucracy; 61 people in Hilo drowned when they apparently failed to hear, or heed, the warning to head for higher ground.

They are among the rarest of natural disasters, and the stealthiest, hiding the energy of a hydrogen bomb in an almost invisible swell. They are as hard to predict as the earthquakes that cause them, which is to say, virtually impossible. Tsunamis actually represent a double uncertainty, because they depend on essentially unknowable factors such as the exact topography of the seabed at the epicenter of a quake. Even watching their seismographs jump last week--even if they could have measured the height of the waves at their origin--geologists couldn't have predicted with any certainty that the sea would rise up in Sri Lanka, a thousand miles away from the fault line. Until then, there had not been a major tsunami in the Indian Ocean since the eruption of Krakatoa in 1883. The last tsunami to cause major damage to the United States came in 1964, when a monster 9.3 earthquake struck Alaska. It sent waves as high as 21 feet at Crescent City, Calif., damaged port facilities as far away as Los Angeles and killed 120 people in three states.

Even local or regional tsunamis, which occur several times a decade, can be deadly. In 1998, an offshore earthquake flooded the coast of Papua New Guinea, killing more than 2,000 and leaving 10,000 homeless. The quake was only a modest 7.1, but it triggered an undersea landslide that sent 45-foot waves crashing onto nearby shores. This attracted the attention of American geologists who study the Cascadia subduction zone, off the coast of Washington and Oregon. Until recently, most scientists believed this fault was unlikely to produce an earthquake powerful enough to threaten the shore. But research over the past decade (including an analysis of tree rings, and historical accounts of a tsunami that might have crossed the ocean and reached Japan) suggest the Cascadia caused a magnitude 9 quake as recently as 300 years ago, and numerous others over the past 7,500 years.

That, by itself, was enough to raise concern, especially in Oregon, which has more than 200,000 people living along its coast, most of them just above sea level. The topography of Washington state and northern California offers a little more protection to most residents. Downtown Seattle is considered relatively safe, given the complicated geography of Puget Sound, but communities such as Bellingham, Port Angeles and Whidbey Island are at risk. Oregon coastal communities "would be inundated" by a powerful tsunami, says Jay Wilson, who heads the state's preparedness program and has the unenviable task of planning for the evacuation of dozens of towns with as little as a half hour's warning. The data from Papua New Guinea gave geologists another reason to worry, because it showed that even a moderate earthquake could set off a tsunami under the right conditions. "A magnitude 7 or 6.5 [well within the range the Cascadia can create] could produce a submarine landslide," says Lori Dengler, a geologist at Humboldt University in northern California. "It could knock out five states. It could be one of the most crippling events the U.S. economy has ever seen."

For that matter, even a repeat of something we know has already happened--the 1964 Aleutian quake, say--could be catastrophic, given the vast increase in coastal populations in the past 40 years. Since seismic data alone can't predict tsunamis, geologists rely on a network of gauges anchored near the shore to measure wave heights all around the Pacific Rim. Since 1996, these gauges have been supplemented by deep ocean sensors, which sit on the ocean floor far from land and measure the pressure of waves passing overhead, transmitting their data by satellite. If these sensors had been in place in the Indian Ocean last week, they could have given a definitive warning of the disaster about to descend on the Subcontinent. If they were working, that is; Dengler notes that of the six already deployed in the Pacific, three were out of service last week.

It's probably small comfort to Los Angeles, but tsunamis are one curse that can't be blamed on the San Andreas Fault. The plate boundaries in southern California are of the slip-fault type, which as a rule do not produce the upthrust earthquakes that give rise to dangerous ocean waves. And residents of the Gulf and Atlantic coasts can rest easy at least until March 16, 2880, when, scientists at the Jet Propulsion Laboratory predict, a 1.1-kilometer asteroid will fall into the Atlantic about 360 miles east of New York, sending a 300-foot-high wall of water crashing into the Eastern Seaboard from Cape Cod to Cape Hatteras. (The event has a probability of three in 1,000, which astronomers consider high.) Unless, that is, the volcanic island of La Palma, in the Canaries, splits apart along a crack that opened after its 1949 eruption. That could send a hundred-cubic-mile chunk of rock sliding into the ocean at 200 miles an hour. Steven Ward, a geophysicist at UC, Santa Cruz, calculates the height of the resulting wave at about 80 feet when it hits the East Coast. "I'd say the slide is pretty ripe," he said last week. "The island is 95 percent ready to go." If it does, the only advice is not to indulge your curiosity. Head for higher ground. And hope that the rest of the world is feeling generous toward the United States.