Recent theories of skill acquisition vary in terms of the specificity of skills. Logan's instance theory suggests that skills are highly specific to experience in that transfer is restricted to situations previously experienced. Anderson's ACT theory proposes that skills are more general, describing how they can apply to situations beyond past experience. Several predictions about the nature of performance improvement on a task, the effect of different forms of training with the task, and transfer to a different version of the task were derived from both theories. These predictions were tested in an experiment that involved subjects solving syllogisms. Although the syllogisms had the same form throughout the experiment, no syllogism was repeated. All of the syllogisms could be solved, however, with the same strategy. Practice resulted in improvement on the task, different training conditions led to different performance strategies, and partial transfer was observed in some conditions. These results were contrary to the predictions derived from the instance theory but were consistent with those derived from the ACT theory. As a result, a number of deficiencies in the instance theory are highlighted. The apparent contradiction between the results of this experiment and recent results reported by Logan is also discussed with a view to reconciliation.