he rivalry between the two teams that decoded the human genome has erupted again, with some academic biologists contending that Celera, their commercial rival, used a method that failed and in fact relied on the academic consortium's method to assemble its own version of the genome.

Celera officials vigorously reject the assertion and say their method worked well.

The dispute is important because it may settle whether the consortium's or Celera's method is the better way of decoding the genomes of other species. Also, if Celera's work was derivative, the consortium's scientists would gather the prizes and glory for decoding the human genome.

The two teams in the fiercely contested race agreed last year to a draw and a truce by announcing in June at a White House ceremony that each had decoded the human genome and by publishing their results at the same time, in rival scientific journals, this February.

But in a series of e-mail messages from February onward, and in a public lecture at Harvard last month, Eric Lander, one of the three principal members of the consortium, has derided the shotgun method championed by J. Craig Venter, Celera's president. Dr. Lander is director of the Whitehead Institute in Boston. The consortium's other two leaders, John E. Sulston of the Sanger Center in England and Robert Waterston of Washington University in St. Louis, pioneered the consortium's method for decoding the human genome.

Celera's whole genome shotgun method "was a flop. No ifs, ands or buts," Dr. Lander wrote in an e-mail message. "Celera did not independently produce a sequence of the genome at all. It rode piggyback" on the consortium's efforts, he said.

Dr. Venter is no less blunt. "We think there is zero legitimacy to anything Eric is saying," Dr. Venter said, "and we don't understand why he is saying it."

The dispute is the latest in a series of increasingly bitter disagreements between the university biologists who are part of the consortium and Dr. Venter, who is a player in both the academic and commercial worlds. The consortium has made its genome sequence free to everyone, denouncing Celera as a commercial operation that will not produce a perfectly finished genome. Celera says its version is better and its subscription cost affordable.

In earlier spats, consortium scientists accused Dr. Venter of a breach of scientific ethics when he announced he would make use of their publicly posted genomic data in completing his version of the human genome. They also tried to prevent Science magazine from publishing Celera's article about the human genome by objecting to Celera's requirements for protecting its data against resale by others.

Celera made no secret of the fact that, to save time and money, it downloaded the consortium's partially assembled data, which was publicly available. It then shredded the data, mixed it with its own and assembled the combined set with its own assembly method. Dr. Lander's charge is that even the shredded public data retained the positional information from the consortium's assembly, without which Celera's assembly method would have failed.

Several other university biologists agree with Dr. Lander's criticisms or believe for different reasons that Celera's method did not work.

"I agree with what Eric is saying and had independently come to the same conclusion," said Philip Green, a computational biologist who wrote two standard programs used by genome researchers. "That shredded data retains the information from the local assemblies, to the extent that you can almost perfectly reconstruct the original sequence."

But Dr. Venter said he shredded the consortium's data specifically to lose its positional information because he suspected the data had been misassembled in places.

The human genome is 3.2 billion chemical units in length, but the machines that figure out their order can handle fragments no larger than 500 units. A genome is decoded by first analyzing the sequence of DNA units in millions of these 500-unit fragments, known as reads, and then having a computer piece them together from their overlaps. In Celera's shotgun approach all the reads are assembled in a single gigantic computation. The consortium, more cautiously, breaks the genome into large chunks, known as BAC clones, whose position on the genome is known, and then assembles each clone from its individual reads.

Eugene W. Myers, a mathematician who is Celera's leading software architect, said Dr. Lander's argument that the shredded data retained the positional information of the BAC clones was disingenuous. "The difficult thing is deciding which overlaps are true and which are false. It's a ridiculous claim that completely ignores the complexity of the problem," he said. His computer program depended only on Celera's data to figure out the main structure of the genome, and used the shredded public data to fill in some gaps within this structure, Dr. Myers said, a step that relied on the coherence of the shredded data but not on its positional information.

As proof that the shotgun assembly method would have worked almost as well without the shredded public data, Dr. Myers says that he has assembled the mouse genome, which is about the same size as the human genome, using Celera-only data, with very similar results.

Gerald M. Rubin, a fruit-fly biologist and vice president of the Howard Hughes Medical Institute, said the only way to assess the truth of Dr. Lander's charge was to find out how the consortium's data had been used in Celera's computer. Those details are not yet available. But the shotgun approach had worked "unequivocally," Dr. Rubin said, in assembling the smaller genome of the fruit fly, a project on which he collaborated with Celera, and he said a successful shotgun assembly of the mouse genome would confirm the method worked with the human genome, too.

The dispute "is an unfortunate continuation of some of the earlier animosity," Dr. Rubin said, but "if Gene Myers is right, Celera is going to end up looking good and Lander will have to apologize."

The dispute over the shredded data is a skirmish in a larger battle, that of whose approach will deserve more credit when the human genome is finally completed. Both sides have produced interim versions of the genome and are now sniping at their rival's.

Dr. Green said of Celera's February report: "More than 20 percent of the genome was not assembled at all or is in tiny scaffolds. That is simply a failure, there is no other word you can use." The shotgun method worked for the fruit fly genome, in his view, but the human genome is much more complicated. "I think basically they could not have done the human genome using a whole genome shotgun, and I think they realized that at some point, which is why they depended so heavily on the public data," he said.

Celera has left itself open to such criticism by the way in which it chose to describe its genome results. In their Science article of February, Dr. Venter and his colleagues say they assembled the human genome in two ways, one by their whole genome shotgun method with shredded public data added, and the other by what they call a "compartmentalized shotgun assembly," a hybrid approach that explicitly drew on the method and data of both Celera and the consortium. But Dr. Venter then proceeded to base all further analysis and gene identification on the hybrid version, ignoring his shotgun version of the genome.

His critics are indignant that the Celera genome analysis rests on the hybrid assembly method instead of on Celera's own — traditionally scientists interpret their own data, not other people's — and scathing about Celera's implied position that the shotgun method worked so well there was no need to prove it.

Dr. Venter said in an interview that the two versions of the genome were very similar but that he had chosen the hybrid method genome for interpretation because it was very slightly more complete. "Our goal was to have the highest quality genome we could. We didn't want to play a silly academic game," he said.

But Celera is now playing that academic game by assembling the human genome with just its own data, Dr. Venter said. The clearest present proof of its shotgun method may lie in its mouse genome, assembled just with Celera data, which the company will offer to its subscribers.

Just how well the shotgun method works is not yet clear, but few critics think it did not work at all, and there is little doubt that Celera accelerated the consortium's timetable. "Would we have the genome sequence now if Celera hadn't been founded? I think the answer is no. I think they made a very serious contribution," Dr. Rubin said.