The training of composite skills requiring differential responding to a lar
ge set of stimuli raises issues about how to break down the whole task into
parts and which parts should be trained first. Components of Morse code re
ception skill were identified, separated, and used to test whether initial
training on a difficult part was more effective than initial training on an
easy part. Initial training on a difficult subset of stimuli and on a diff
icult subtask both yielded disadvantages rather than the advantage implied
by recent findings with different tasks. Incremental training should begin
with the part yielding the most effective strategic skills, which appear to
depend on characteristics of the task. In both present experiments, easy i
nitial training led to adoption of an effective unitization strategy for re
presenting codes. The hypothesis that procedural reinstatement at delayed t
esting leads to better retention was supported and extended.