A senior congressional investigator has accused his agency of covering up a scientific fraud among builders of a $26 billion system meant to shield the nation from nuclear attack. The disputed weapon is the centerpiece of the Bush administration's anti-missile plan, which is expected to cost more than $250 billion over the next two decades. The investigator, Subrata Ghoshroy of the Government Accountability Office, led technical analyses of a prototype warhead for the anti-missile weapon in an 18-month study, winning awards for his "great care" and "tremendous skill and patience." Ghoshroy now says his agency ignored evidence that the two main contractors had doctored data, skewed test results and made false statements in a 2002 report that credited the contractors with revealing the warhead's failings to the government. The agency strongly denied his accusations, insisting that its anti-missile report was impartial and that it was right to exonerate the contractors of a cover-up. The dispute is unusual. Rarely in the 85-year history of the GAO, an investigative arm of Congress with a reputation for nonpartisan accuracy, has a dissenter emerged publicly from its ranks. And Ghoshroy's assertions raise new questions about the Boeing Co.'s military arm, the main contractor for the troubled $26 billion system of interceptor rockets now being installed in Alaska and California. The system's "kill vehicles" are to zoom into space and destroy enemy warheads by force of impact. But years of test failures have thrown the program into disarray, and the military has recently begun to look for a kill vehicle of greater reliability. Ghoshroy, 56, a senior analyst with seven years of service at the accountability office, makes his charges in a recent letter to Rep. Howard L. Berman, a California Democrat, who, along with Sen. Charles E. Grassley, an Iowa Republican, originally requested the GAO study. Berman's office provided his letter to The New York Times, along with dozens of Ghoshroy's documents. After the GAO made its report public in February 2002, Ghoshroy quietly tried to have his agency reverse itself and grew increasingly frustrated. He took a sabbatical at Harvard. Ghoshroy is now on leave from the accountability office as a research fellow at the Massachusetts Institute of Technology. In an interview, David M. Walker, the head of the GAO, called the senior analyst "a relatively low-level, disgruntled employee" out of step with his peers. He denied that his agency had produced a biased report and defended the rigor of its investigations. "We don't pull any punches," Walker said. "It's almost laughable for anybody to say that." Pentagon planners hail the weapon, known as the ground-based midcourse interceptor, as a hedge against disaster. Skeptics ridicule it as an unworkable defense against a nonexistent threat. The dispute over its reliability began a decade ago. Nira Schwartz, a senior engineer in 1995 and 1996 at the military contractor TRW, told her superiors that the company had falsified research findings meant to help kill vehicles differentiate incoming warheads from clouds of decoys. In April 1996, Schwartz filed a suit under the False Claims Act, a federal law that allows heavy fines against contractors who lie about their government work. TRW denied her accusations. She subsequently singled out the prototype kill vehicle's first flight test, in June 1997, arguing that the contractors falsified data from it. The flight cost $100 million. TRW was a Boeing subcontractor. Boeing, in turn, was competing against other companies to build the overall kill vehicle. Both denied any impropriety. In 2000, Grassley and Berman asked the GAO to examine Schwartz's charges. Ghoshroy became the main technical analyst. Almost immediately, he recalled, the GAO team found signs of a cover-up - for instance, disturbing charts buried at the back of an upbeat report. On April 12, a GAO manager wrote a draft summary of the team's findings. It strongly backed Schwartz, saying the contractors had "excluded some data and modified statistical techniques." Around this time, the GAO team was directed to switch its focus from looking for fraud to searching for contractor admissions of failure. While it found written reports that disclosed some flight-test problems before Schwartz revealed them to federal investigators, it was unable to document them all. Finally, the team learned of a meeting in late August 1997 at which contractor personnel had reportedly made complete oral disclosures. But no contractor or federal official could recall anything specific about this meeting - no date, place, agenda or list of attendees. Ghoshroy came to believe that the meeting had never happened, he said. Even so, the GAO report incorporated its claims. After the report was made public, Ghoshroy in private called for an independent investigation of its integrity. The GAO conducted three internal inquiries, each absolving the agency of any wrongdoing. In an interview and written responses to questions, Walker of the GAO said part of the controversy grew from a "communication breakdown" with Berman's office over the report's scope. Walker emphasized that the strong consensus at the GAO was against Ghoshroy. And he defended the report's use of the disputed oral disclosures, saying it relied "more heavily" on written admissions. The contractors never presented the exonerating GAO study in court. In early 2003, a federal judge threw out Schwartz's suit after deciding that going ahead would release military secrets that could hurt national security. Boeing declined to comment on the dispute.