Selecting conformations for side-chains is an important subtask in building three-dimensional protein models. Side-chain placement is a difficult problem because of the large search space that has to be explored. We show that the side-chain placement problem can be expressed as a CLP program in which rotamer conformations are used as values for finite domain variables, and bad atomic clashes involving rotamers are represented as constraints. We present a new side-chain placement method that uses a series of automatically generated CLP programs to represent successively tighter side-chain packing constraints. By using these programs iteratively our method produces side-chain conformation predictions whose accuracy is comparable with that of other methods. The resulting system provides a testbed for evaluating the quality of protein models obtained using different domain enumeration heuristics and side-chain rotamer libraries.