A statistical mechanics description of complex molecular systems involves the determination of ensembles of conformations that represent their Boltzmann distributions. The observable properties of these systems can be then predicted by calculating averages over such ensembles. In principle, given accurate energy functions and efficient sampling methods, these ensembles can be generated by molecular dynamics simulations. In practice, however, often the energy functions are known only approximately and the sampling can be carried out only in a limited manner. We describe here a method that enables to increase simultaneously both the quality of the energy functions and of the extent of the sampling in a system-dependent manner. The method is based on the incorporation of experimental data as replica-averaged structural restraints in molecular dynamics simulations and exploits the metadynamics framework to enhance the sampling. The application to the case of α-conotoxin SI, a 13-residue peptide that has been characterized extensively by experimental measurements, shows that the approach that we describe enables accurate free energy landscapes to be generated. The analysis of these landscapes indicates the presence of a low population state in equilibrium with the native state in which the only aromatic residue of α-conotoxin SI is exposed to the solvent, which is a feature that may predispose the peptide to interact with its partners.