T here's no exact number or good guess as to how many "poor performers" work in the federal government. We're talking about the deadwood -- those who don't pull their own weight.

In 1999, supervisors estimated that 3.7 percent of the federal workforce fell into that category, according to a survey by the Office of Personnel Management.

In 2000, employees suggested that 14.3 percent of their co-workers were not performing at reasonable levels, a survey by the Merit Systems Protection Board found.

A year later, a survey by Brookings Institution fellow Paul C. Light found that, on average, federal employees believed that 23.5 percent of their colleagues were "not up to par."

The disparity in responses is probably related to the way the questions were asked. Many senior officials think that problem employees are relatively few, considering that the government has about 1.8 million civil service employees.

(In fiscal 2003, data compiled by OPM showed that 99.7 percent of federal employees received a performance rating of "acceptable" or higher, such as "fully successful," "exceeds expectations" and "outstanding.")

But there is widespread agreement that a few problem employees usually cause big trouble in offices and, as the saying goes, poison the well.

The issue of poor performers was recently reviewed by the Government Accountability Office at the request of Republican leaders of the House Government Reform Committee. The review seems timely, because federal agencies are being urged by the Bush administration to tighten up their systems for rewarding the best workers and weeding out the bad workers.

At two departments, Defense and Homeland Security, officials are planning to replace the General Schedule, which provides predictable pay raises, with systems that link raises to job performance and reduce employees' appeal time.

At Defense and Homeland Security, the revamped personnel systems make it possible for supervisors to give little or no pay raise to poor performers or, under certain conditions, even give them a pay cut.

In its letter, GAO told the House committee that various reports and surveys show agencies have been reluctant to take on problem employees for numerous reasons. Some supervisors think the rules to correct performance are too complex and take up too much time. Some feel they have not been trained in performance management. Some opt to avoid the issue because they don't like workplace confrontations.

Management associations also contend that supervisors and executives risk grievances and complaints alleging discrimination from employees who try to get even and intimidate management. There's also a perception in some agencies that high-level management will not support supervisors trying to fire problem employees.

GAO said that most agencies are better off if they try to address poor performance sooner rather than later. Effective management systems help supervisors provide candid and constructive feedback aimed at helping employees improve, GAO said.

In addition, GAO said, agencies need to deal with problem employees while they are on probation -- usually their first year on the job -- and before they are eligible for full job protections. Defense and Homeland Security have the option of extending probation beyond the standard one-year period as part of their personnel overhauls, GAO noted.

Pentagon officials have pointed to their "demonstration projects" -- which test alternative pay and personnel rules -- as one of the reasons they are eager to overhaul Defense rules that cover more than 700,000 employees.

The personnel projects appear to have encouraged employees to work harder, according to a 2002 OPM evaluation of the projects carried out at a dozen science and research labs.

"The flexibility to pay higher starting salaries has been helpful and the labs are retaining more of their top performers," the OPM report said. "There also has been a positive effect . . . on motivation and willingness to work harder since the implementation of pay for performance."

In some Defense labs, the decision to link pay raises to job performance ratings appears to have encouraged the slugs in the workforce to pack up and move on. "Turnover was significantly lower among highly rated employees and higher among employees with lower ratings," OPM found.

E-mail: barrs@washpost.com.