The purpose of this paper is to examine what is known about the costs and cost-effectiveness of computer-assisted instruction (CAI) by presenting an appropriate cost methodology, applying it to CAI, and exploring the validity of four popular assumptions. These assumptions are: (1) that computer hardware accounts for most of the cost of delivering CAI; (2) that drastic declines in future costs of computers will create similar reductions in CAI; (3) that networks of microcomputers used for CAI are less costly than minicomputers with similar capabilities; and (4) that CAI has been found to be more cost effective than other instructional alternatives. The first three assumptions are directly contradicted by the available evidence. With respect to cost-effectiveness in raising student achievement in mathematics and reading, CAI was found to be more cost-effective than reducing class size, increasing the length of the school day, or adult tutoring, but considerably less cost-effective than peer tutoring. A bibliography of 11 citations is appended. (Author/MLF)