This study investigates how stimulus complexity influences human learning of a Sequence Learning (SL). A usual approach to analysis of learning data is based on a learning curve (LC), which shows how a probability of correct response on each step changes across an SL. Usually, a single logistic curve (LgC) provides a good approximation of LCs under various conditions. However, our results show that increase in complexity of a stimulus causes usage of more than one LgC for approximation. In the case of high complexity, LgC curve no longer describes a learning dynamics, but only its general trend. We also extracted other functions needed to approximate a learning dynamics under high complexity. Furthermore, to approximate learning dynamics under higher complexity, we have to use the functions employed under the lower complexity conditions and some additional functions. Therefore we hypothesize a hierarchical structure of the learning process related with stimulus complexity.