Arguing that the most widespread argument in favor of the teaching of programming concerns its possible impact on generalizable cognitive skills, this paper addresses the "how" of transfer. The outlines of a theory of the mechanisms of transfer are presented, the theory is used to examine the contrasts between certain studies that did and did not obtain positive transfer results from programming, and a discussion is presented of the kinds of transfer that can be expected from programming and when. Two roads to transfer are identified (high and low) and examples of each are provided. In addition, six broad categories of transfer that might occur with programming are described: (1) mathematical and geometric concepts and principles; (2) problem solving, problem finding and problem management; (3) abilities of formal reasoning and representation; (4) models of knowledge, thinking, and learning; (5) cognitive styles; and (6) enthusiasms and tolerances. It is concluded that although programming instruction can improve cognitive skills under the right conditions, implementing such conditions on a wide scale may be difficult, and programming will have to compete in the intellectual and economic markets with a number of other approaches to the same general problem. A five-page list of references is provided. (JB)