Abstract The paper investigates a novel approach, based on Constraint Logic Programming (CLP), to predict potential 3D conformations of a protein via fragments assembly. The fragments are extracted and clustered by a preprocessor from a database of known protein structures. Assembling fragments into a complete conformation is modeled as a constraint satisfaction problem solved using CLP. The approach makes use of a simpli?ed C?-side chain centroid protein model, that offers ef?ciency and a good approximation for space ?lling. The approach adapts existing energy models for protein representation and applies a large neighboring search (LNS) strategy. The results show the feasibility and ef?ciency of the method, and the declarative nature of the approach simpli?es the introduction of additional knowledge and variations of the model.