The present investigation was concerned with the variability-of-practice hypothesis of Schmidt's schema theory of motor learning; i. e. it was sought to determine if there is an optimal way to structure the variable practice session with regard to schema formation. Furthermore, children's motor recognition (movement evaluation) was examined, in addition to recall (movement production), since apparently no such studies had been done within the context of schema theory before. Eighty-eight girls and boys with a mean age of 11.3 years were tested on a throwing task. Three groups of subjects practised under variable conditions (different target distances and weights) that differed with regard to contextual interference and sequential organization of task variations; another group received constant throwing practice. Following practice trials with knowledge of results, the four treatment groups and a control group without prior practice were required to produce and evaluate a novel response (novel target distance and weight) without vision and knowledge of results. The results supported the variability-of-practice hypothesis, in that variable practice in general facilitated recall and recognition on the novel task. Furthermore, variable practice with the highest level of contextual interference (random order of tasks) produced most effective transfer performance, suggesting that the structure of the variable-practice session in fact plays a role with regard to the effectiveness of motor schema development. Contrary to the ‘organization hypothesis’ (Gentile and Nacson, 1976) a hierarchical organization of movement variations was not more beneficial to schema formation than random practice.