When acting in non-deterministic environments, autonomous agents must balance between long-term, complex goals with unpredictable events and reactive behavior. In this context, hierarchical operational models are attractive in that they allow the execution of complex behavior either in a purely reactive fashion or guided by a planning process. Just like for HTN models with which they share most characteristics, one key bottleneck in the exploitation of operational models is their acquisition. In this paper, we introduce an algorithm for learning hierarchical operational models from a set of demonstrations. Given an initial vocabulary of tasks and some demonstrations of how they could be achieved, we present how each task can be associated to a set of methods capturing the operational knowledge of how it can be achieved. We present the structure of the learned models, the algorithm used to learn them as well as a preliminary evaluation of this algorithm.