AT FIRST SIGHT, the Bush administration's proposal to reform pay scales for civil servants appears common-sensical. There's no reason why pay increases should be uniform, as they currently are; they should reward performance and reflect the job market. If private-sector wages are rising quickly in San Francisco or Houston, the government has to pay more in those cities to retain people; if wages for computer systems specialists are rising fast, the government must offer a premium for those skills. As the administration rightly says, no private company awards raises to people just for showing up. If the government wants to use taxpayers' money effectively, it needs a pay scale designed to retain the right kinds of workers.

The trouble is that implementing a different pay system is not going to be simple. Before you can link salary to performance, performance has to be defined -- and this isn't easy. Companies can judge the performance of a business unit by how much money it brings in. But government departments can't be measured by that yardstick. In some areas, alternative performance benchmarks are obvious: You can judge a customer service official at Medicare or the Department of Veterans Affairs by how many calls he answers per hour and how often his answer satisfies the caller. But some federal functions don't have a quantifiable output. The international staff of the U.S. Treasury, for example, is supposed to monitor developments in the international economy, coordinate with other governments and occasionally push a policy idea. How do you measure its "output"?

A good manager can get around this problem, judging subordinates on the quality of their work even where objective and quantitative measures are impossible. But that requires giving managers the discretion to reach negative findings and -- just as important -- an incentive to do so. In rolling out its proposal, the administration has stressed that the plan involves no dilution of union power or the right of employees to appeal management decisions. But if employees can appeal, what manager is going to hand out negative appraisals on the basis of necessarily subjective judgments? It's not clear that civil service managers, who are paid less than their private-sector peers, have the necessary incentives to make hard personnel decisions.

This is not to say that no reform should be attempted. The administration promises to spend "hundreds of millions" on training managers to make tough but valid judgments, and its reform is supposed to tie managers' pay to their willingness to make these calls; in any case, moving the civil service toward incentive pay, however imperfectly, is better than doing nothing.

But the fact remains that reform will be hard, and therefore a gradual, piecemeal approach may be better than an ambitious sweep. The administration's proposal would allow agencies to reform at slightly different rates, and that flexibility should be enhanced by pushing the deadline for full implementation further out than 2010, the current target. Better to move cautiously, learning lessons along the way, than to rush into a costly and disruptive reform across the whole civil service.