A widely-acclaimed cancer researcher at Harvard Medical School is under investigation a second time for "possible discrepancies in test results" conducted by his research laboratory.

Eric T. Fossel, an associate professor of radiology at Beth Israel Hospital, is being investigated by the hospital and the Medical School for allegedly tampering with the data for research he conducted on developing a new technique for cancer detection. The story was first reported in the Boston Globe yesterday.

A Beth Israel inquiry into Fossel's work several years ago turned up no evidence of actual research fraud, according to the Globe and a source familiar with the investigation. But the hospital determined then that Fossel's conclusions were "not strictly valid," the source said.

In the statement, Beth Israel said it was "made aware from an external funding source" on Nov. 2 of alleged problems with Fossel's research.

Beth Israel's president, Dr. Mitchell T. Rabkin, did not return a phone call yesterday.

Fossel could not be reached for comment. According to an associate, the 51-year-old researcher was scheduled to deliver a major address at the 14th Annual San Antonio Breast Cancer Symposium in Texas on Saturday. But Fossel, who is reported to have undergone major heart surgery recently, was notregistered at the Hyatt Regency hotel, where theconference was held.

In a statement, Beth Israel confirmed theinvestigation but did not specify the nature ofthe alleged discrepancies in Fossel's work. Theprofessor's laboratory will remain open pendingthe results of the inquiry, the statement said.

According to the Globe and sources familiarwith the investigation, Fossel is alleged to haveused deliberately misleading data in reaching hisconclusions. Those results, prominently featuredin the New England Journal of Medicine in November1986, received widespread attention upon theirrelease.

The 1986 article suggested that a widely usedtechnology for examining molecular structurescould be effective in screening patients forotherwise undetectable cancers.

According to the Fossel study, the test, knownas nuclear magnetic resonance spectroscopy (NMR),identified changes in the molecular structure ofoncolipids--fatty acids in the blood plasma thatare thought to be associated with cancer.

"We conclude that these preliminary resultsdemonstrate that...NMR spectroscopy is apotentially valuable approach to the detection ofcancer and the monitoring of therapy," the studysaid.

In essence, according to cancer experts, Fosselseemed to have discovered a fingerprint forcancer. The study was enthusiastically welcomed bythe research community.

"The work was very exciting at the time," saidone cancer researcher at Beth Israel, who spoke oncondition of anonymity.

But a follow-up study that was published in theJournal four years later, in April of 1990, failedto reproduce Fossel's results. Indeed the study,which was conducted by a team of scientists inNorway, reached exactly the opposite deduction.

"We conclude that...NMR spectroscopy is notgenerally reliable for the detection of cancer,"the 1990 report said.

Other researchers also were unable to reproduceFossel's results. According to Dr. Paul Okunieff,chief of radiation oncology at the National CancerInstitute, in Maryland, even Fossel hasn't beenable to reproduce the results.