The human ability to fast and efficiently learn via instructions is remarkable. Initially encoded as declarative stimulus-response (S-R) rules, instructions need to be converted into a procedural format for successful execution. At the stage of rule implementation, it is usually difficult to disentangle these two levels of rule representation. We compared direct instructions (‘if Stimulus A, press Key T’) to indirect instructions (‘if Stimulus B, do not press Key T but another Key’) to explore whether this approach could be a valuable means to achieve distinct measures of declarative and procedural representations, respectively. To this end, following these different instruction types, we assessed behavioral performance during repeated direct rule implementation as well as during a subsequent rule recognition test. We found that compared to direct instructions implementation response time (RT) was significantly longer on indirectly instructed rules at all repetitions. Furthermore, a positive correlation between implementation RT and rule recognition accuracy was only observed on indirectly instructed rules, which suggests the continued importance of the instruction-related declarative representation for that instruction type. At the same time and at constant error rates, the tendency to alternate between response options (i.e., switch probability) was decreasing over repetitions for indirectly but not for directly instructed rules, indicating the emergence of an implementation-related procedural rule. Importantly, S-R rule identities at the declarative and procedural level differ for indirectly instructed rules but not for directly instructed ones. Therefore, our results indeed provide a basis for distinguishing between representation types, potentially also at the neural level.