Implicit learning is thought to be one basic learning mechanism, taking place without conscious awareness about the fact that one learns or what one learns. Implicit learning is assumed to occur in distributed sets of encapsulated, specialized, and parallel working modules. However, it is not clear yet what exactly is processed in these modules. While there is ample evidence for a distinction between modalities (e.g., visual, auditory perception), the representational content of the modules might even be distinguished by features within the same modalities (e.g., color, and shape within the visual modality). One way to distinguish between these views is to study cross-modal learning (e.g., learning of a color-shape sequence within the visual modality). According to modularized theories, the encapsulated modules might not exchange their sequence knowledge which would prevent cross-modal learning if the modules process single features. Indeed, we showed in Experiment 1 that participants did not learn a cross-modal sequence of colors and shapes (e.g., red – blue – triangle) but an abstract sequence of categories (e.g., color – color – shape). An alternative assumption is that participants did not learn cross-modal transitions because the task set in implicit learning tasks might stress the unimodal transitions. In Experiment 2, we stressed the importance of cross-modal sequence transitions in an induction phase. The findings showed that the task set influenced whether the participants learned the concrete or the abstract cross-modal sequence. We conclude that implicit learning is based on the processing of features and the current task set.