Hierarchical Task Networks (HTNs) are a common formalism in automated planning. However, HTN models are mostly designed by hand by expert users. While many of the state-ofthe-art approaches for learning HTN try and learn the structure and its parameterization in a single step, other focus specifically on learning the structure of the model. Many of these structure-focused approaches, however, learn models with non-parameterized actions, task and methods, which limits their generalization capabilities. In this paper, we propose a constraint satisfaction-based approach for extracting parameters for a given HTN structure using a set of demonstration traces.