The best evidence on effective instruction that we can garner is fragile, often stripped of any external validity. Almost every curriculum scheme works in some sites, and none has ever worked well everywhere. That has been the finding time after time when state and federal educational initiatives have been evaluated. There is a useful role for research in improving the teaching of reading. But that role is limited because the research available is limited in many ways. Thus a federal campaign to impose particular one-size-fits-all instructional materials and methods on America's children and their teachers is neither scientific nor productive. While the author's views occupy something of a middle ground in the debate over the report of the National Reading Panel, he believes that the federal government should not be setting policy about reading instruction. He offers recommendations that, if followed, may increase our chances of leaving no child behind.