Instructions have long been considered a highly efficient route to knowledge acquisition. To enable actual instruction implementation, initial declarative representations are often assumed to be transformed into actionable procedural ones. We employed different study designs that allowed us to examine the influence of learning intention, instruction repetition, as well as working memory load and span. Moreover, we tested the hypothesis that declarative-procedural transformation should be bound to a specific response modality and not be transferable across different modalities. First, our results suggest that instruction-based learning involves active intention to learn. Second, compared to trial-and-error learning, repeated instruction implementation accelerates short-term automatization while saving on working memory resources. Third, procedural encoding is modality-specific as indicated by an implicit priming test. This contrasted with explicit test performance, serving as a marker of declarative encoding, which was independent of modality transition and uncorrelated with implicit test performance. Fourth, we found that procedural encoding occurs also in the absence of overt instruction implementation. Fifth and surprisingly, both implicit and explicit test scores benefited from previously non-implemented instructions compared to previously implemented instructions. Overall, our results provide clues how to learn efficiently via explicit instructions.