Characterizing the inherent flexibility of a protein provides an important link between structure and function. In this article, we examine some of the methods used to determine protein flexibility, and address several unanswered questions relating to them. We perform 4 ns simulations of bovine pancreatic trypsin inhibitor (BPTI) in solution and in a crystal. For comparison, we also calculate atomic fluctuations from room temperature x-ray diffraction data by two different methods: single copy refinement with isotropic B-factors, and constrained multiple copy refinement. We reach the following conclusions: (1) Crystal contacts significantly reduce atomic fluctuations, especially in the flexible loop regions. (2) Center of mass motion in the crystal contributes 0.1-0.2 Angstrom to the rms fluctuations, with little variation between parts of the protein. (3) Isotropic B-factors are an accurate measure of atomic motion in the stable parts of the protein, but significantly underestimate motion of the flexible sidechains. (4) Nanosecond scale simulations can obtain a reasonable sampling of backbone atomic motion in the most stable regions, but are still too short to allow flexible regions to explore their full range of motion.