The U.N.-backed panel investigating fraud in Afghanistan's presidential vote has agreed to allow a recount of just a sampling of hundreds of thousands of suspect ballots to speed a process some fear could stretch into spring, an official said Monday.

Preliminary results from the Aug. 20 poll show President Hamid Karzai winning with 54.6 percent, but if enough votes are thrown out because of fraud it could drop him below 50 percent, forcing a runoff with top challenger Abdullah Abdullah.

Time for a second vote is running short: Winter snows start early and make much of the country impassable. But waiting for spring to hold the poll could create a power vacuum in a country already destabilized by the resurgent Taliban and delay both government projects and military operations.

The U.N.-backed Electoral Complaints Commission, the final arbiter on disputed ballots, has said there was "clear and compelling evidence of fraud" countrywide and ordered a recount and audit of ballots at about 3,000 polling stations because of suspect tallies. It is unclear how many ballots that includes, but European Union monitors have said enough Karzai votes were questionable that he could drop below the benchmark.

Afghan officials have said a full audit and recount could take two or three months. The complaints panel is meeting with Afghan election officials and outside election experts to design a sample size that will be accurate enough to determine fraud but speed up the procedure, said Grant Kippen, the Canadian head of the panel.

"We will save time using the sample approach versus going out and looking at every single ballot box," Kippen said. While tally sheets have arrived in Kabul, the actual ballot papers are stored in the provinces and retrieving them securely is not a quick process.

Kippen said that the sampling process should still give an accurate representation of the vote, comparing it with opinion polls that reach a high degree of accuracy by asking questions of a portion of the population.

"You're making certain judgments based on a statistically relevant sample of a population, and those have proved to be highly accurate," he said.