Considering that computers have been with us for some time, I am surprised at the primitive state of our critical capability to judge their usefulness in educational settings. What I see often are arguments at very general levels: "Computers are dynamic and interactive and can respond patiently in an individualized manner." To be sure, one can rely to some extent on such generalizations, but most confidently in retrospect rather than in anticipation of good educational outcomes. Engineering and science are driven by expectations and promises like these, but anyone who has engaged in design or science knows how these can fail in the details. For some reason the promises just don't "get down to brass tacks;" general goals and possibilities don't really tell us what to do. Critical judgments of particular software are also too often driven by face validity. It is as if one can see quickly and easily the features that lead to in-depth learning. One hears that the software contains a lot of information; it offers the user immense freedom; the simulation is dynamic and interactive; ideas are rendered concrete and manipulable; the program satisfies a checklist of "good educational software" rules, and has a clever user interface.