Magill and Hall (1990) hypothesized that the contextual interference (CI) effect is found only when task variations to be learned are governed by different generalized motor programs (GMPs). The present experiments examined their hypothesis by requiring subjects to learn variations of a tapping task that had either different (Experiment 1) or the same (Experiment 2) relative timing structure. In each experiment, subjects (N = 36) performed 270 acquisition trials with knowledge of results (KR) in either a blocked or a serial order. One day later, subjects performed 30 retention trials without KR. In data analyses, errors due to parameter modifications were dissociated from errors due to GMP construction to examine which process was responsible for the CI effect. In both experiments, parameter learning created a CI effect while GMP learning failed to produce a CI effect. In the light of these findings, a modification is proposed to the Magill and Hall (1990) hypothesis that takes into account these distinct processes in motor learning.