Several obstacles were encountered and overcome during the structure determination of human cytomegalovirus protease. Dehydration of crystals, by exposing them to higher concentrations of the precipitant, reduced the mosaicity of the crystals and may have also resolved their microscopic twinning. The initial phase information was obtained with the selenomethionyl multiple-wavelength anomalous diffraction technique. However, site-specific mutagenesis was required to introduce extra Met residues into the protease. The phase information had to be improved by non-crystallographic symmetry averaging, initially among three 'crystal forms'. A change in the composition of the artificial mother liquor led to a significant improvement, from 3.0 and 2.0A resolution, in the diffraction quality of the crystals. The experiences reported here may prove useful to structure determination of other proteins.