Time has run out for a researcher's prediction that a large earthquake would hit California by 5 September. Vladimir Keilis-Borok had predicted that such an event would occur at some point in the past nine months. But, fortunately for residents, nothing of the kind happened.

Keilis-Borok, a geophysicist at the University of California, Los Angeles, hasn't given up. “We can use the mistakes to improve our methodology,” he says.

Keilis-Borok has had some success in the past. In 2003, his team made forecasts that were fulfilled by the Tokachi-oki earthquake in Japan on 26 September, and the Californian San Simeon earthquake on 22 December.

Recently he predicted that an earthquake of magnitude 6.4 or greater would hit somewhere in 32,000 square kilometres of southern California (see map). The United States Geological Survey and the California Earthquake Prediction Evaluation Council said the method “appears to be a legitimate approach”. But they took no action; a nine-month window cannot be used to make decisions about evacuation.

Keilis-Borok looks for patterns in the smaller earthquakes that precede large ones. He uses a similar method to analyse other systems such as crime waves and elections; his numbers predict that George W. Bush will win in November.

But some experts think Keilis-Borok's successes have been down to luck. Based on earthquake history, there is a 30% chance of one hitting Tokachi-oki during any nine-month period, although there is only a 2–5% chance of one in San Simeon. Tom Jordan, head of the Southern California Earthquake Center in Los Angeles, says Keilis-Borok's method uses complex algorithms that will have to be more thoroughly evaluated before critics are won over.