The numbers are still relatively small, and they do not mean that the virus has mutated to pass easily between people — a change that could touch off a worldwide epidemic. All the clusters of cases have been among relatives or in nurses who were in long, close contact with patients.

But the clusters — in Indonesia, Thailand, Turkey, Azerbaijan, Iraq and Vietnam — paint a grimmer picture of the virus's potential to pass from human to human than is normally described by public health officials, who usually say such cases are "rare."

Until recently, World Health Organization representatives have said there were only two or three such cases. On May 24 Dr. Julie L. Gerberding, director of the federal Centers for Disease Control and Prevention in Atlanta, estimated that there had been "at least three." Then, last Tuesday, Maria Cheng, a W.H.O. spokeswoman, said there were "probably about half a dozen." She added, "I don't think anybody's got a solid number."

And Dr. Angus Nicoll, chief of flu activities at the European Center for Disease Prevention and Control, acknowledged that "we are probably underestimating the extent of person-to-person transmission."

The handful of cases usually cited, he said, are "just the open-and-shut ones," like the infections of nurses in the 1997 Hong Kong outbreak and of a Bangkok office worker who died in 2004 after tending her daughter who fell sick on an aunt's farm.

Most clusters are hard to investigate, he said, because they may not even be noticed until a victim is hospitalized, and are often in remote villages where people fear talking. Also, he said, by the time doctors from Geneva arrive to take samples, local authorities "have often killed all the chickens and covered everything with lime."

The W.H.O. is generally conservative in its announcements and, as a United Nations agency, is sometimes limited by member states in what it is permitted to say about them.

Still, several scientists have noted that there are many clusters in which human-to-human infection may be a more logical explanation than the idea that relatives who fell sick days apart got the virus from the same dying bird.

For example, in a letter published last November in Emerging Infectious Diseases analyzing 15 family clusters from 2003 through mid-2005 in Southeast Asia, scientists from the disease control centers, the W.H.O. and several Asian health ministries noted that four clusters had gaps of more than seven days between the time family members got sick. They questioned conventional wisdom that only one, the Bangkok office worker, was "likely" human-to-human.

In one Vietnam cluster, not only did a young man, his teenage sister and 80-year-old grandfather test positive for A(H5N1) avian flu, but two nurses tending them developed severe pneumonia, and one tested positive.

In another questionable case, the Vietnamese government's assertion that a man developed the flu 16 days after eating raw duck-blood pudding was publicly ridiculed by a prominent flu specialist at Hong Kong University, who said it was more likely that he got it from his sick brother.

Dr. Henry L. Niman, a biochemist in Pittsburgh who has become a hero to many Internet flu watchers and a gadfly to public health authorities, has argued for weeks that there have been 20 to 30 human-to-human infections.

Dr. Niman says the authors of the Emerging Infectious Diseases article were too conservative: even though the dates in it were fragmentary, it was possible to infer that in about 10 of the 15 cases, there was a gap in onset dates of at least five days, which would fit with the flu's incubation time of two to five days.

And in a study published just last month about a village in Azerbaijan, scientists from the W.H.O. and the United States Navy said human-to-human transmission was possible. That conclusion essentially agreed with what Dr. Niman had been arguing since early March — that it was unlikely that seven infections among six relatives and a neighbor, with onset dates stretching from Feb. 15 to March 4, had all been picked up from dying wild swans that the family had plucked for feathers in a nearby swamp in early February.

While Dr. Niman is an irritant to public health officials, his digging sometimes pushes them to change conclusions, as it did in the recent Indonesia case. The W.H.O. at first said an undercooked pig might have infected the whole family, but Dr. Niman discovered that the hostess of the barbecue was sick two days before the barbecue and the last relative was infected two weeks after it.

His prodding, picked up by journalists, eventually led the W.H.O. to concede that no pig was to blame and that the virus probably had jumped from human to human to human.

The health organization's periodic updates on the number of avian flu cases and the death toll concentrate on cases confirmed by laboratories. The updates use no names and are often cleared by the affected country's health minister.

Dr. Niman, by contrast, trolls local press and radio reports and uses Google software to translate them — sometimes hilariously — looking for family names, onset dates and death dates.

For example, a May 15 report quotes a village midwife named Spoilt describing the death of a woman in Kubu Sembilang, Indonesia and the hospitalization of one of her sons:

"Praise br Ginting experienced was sick to last April 27 2006, with the sign of the continuous high fever to the temperature of his body reached 390 C was accompanied by coughs... Added Spoilt, second casualties Roy Karo-Karo that also the son of the uterus from Praise br. Gintin after his mother died last May 3, also fell ill, afterwards was reconciled to RSU Kabanjahe."

Dr. Niman contends that the largest human-to-human cluster so far was not in Indonesia, but in Dogubayazit, Turkey, in January. W.H.O. updates recorded 12 infected in three clusters, and quoted the Turkish Health Ministry blaming chickens and ducks. Dr. Niman counted 30 hospitalized with symptoms and said the three clusters were all cousins with the last names of Kocyigit and Ozcan, and that most fell sick after a big family party on Dec. 24 that was attended by a teenager who fell sick on Dec. 18 and died Jan. 1.

A patriarch, Dr. Niman said, told local papers that the two branches had had dinner together six days after the 14-year-old, Mehmet Ali Kocyigit, had shown mild symptoms. He died on Jan. 1, and several other young members of the two families died shortly after, with other relatives showing symptoms until Jan. 16. No scientific study of that outbreak has been released.

Dr. Niman also said clusters were becoming more frequent, especially in Indonesia. Just last week two more emerged there, one including a nurse whose infection has not yet been confirmed. With 36 deaths, Indonesia is expected to eclipse Vietnam soon as the world's worst-hit country.

Dr. David Nabarro, chief pandemic flu coordinator for the United Nations, said that even if some unexplained cases were human-to-human, it does not yet mean that the pandemic alert system, now at Level 3, "No or very limited human-human transmission," should be raised to Level 4, "Increased human-human transmission."

Level 4 means the virus has mutated until it moves between some people who have been only in brief contact, as a cold does. Right now, Dr. Nabarro said, any human transmission is "very inefficient."

Level 6, meaning a pandemic has begun, is defined as "efficient and sustained" human transmission.

Ms. Cheng of the W.H.O. said that even if there were more clusters, the alert would remain at Level 3 as long as the virus dies out by itself.

"A lot of this is subjective, a judgment on how efficiently the virus is infecting people," she said. "If it becomes more common, we'd convene a task force to raise the alert level."