The structure and dynamics of biological molecules are essential for their function. Consequently, a wealth of experimental techniques have been developed to study these features. However, while experiments yield detailed information about geometrical features of molecules, this information is often incomplete and subject to averaging through both space and time. In addition, experimental noise often is signicant. These facts complicate the use of the information in the construction of models representing the conformational properties of the molecules.
In this thesis I review the current literature in the eld of protein structure and structural ensemble determination from experimental data. Following this I present three original research paper which address a number of current shortcomings in the eld. Firstly, a method to increase the efficiency of probabilistic structure determination is presented. Second, a generalization of this structural inference framework is presented, to account for exibility in the molecules. Finally, I apply the generalization to restrain a simulation of the native uctuations of a protein using experimental data.