The noble gas krypton is shown to bind to crystallized proteins in a similar way to xenon [Schiltz, Prange & Fourme (1994). J. Appl. Cryst. 27, 950-960]. Preliminary tests show that the major krypton binding sites are essentially identical to those of xenon. Noticeable substitution is achieved only at substantially higher pressures (above 50 x 10(5) Pa). As is the case for xenon, the protein complexes with krypton are highly isomorphous with the native structure so that these complexes can be used for phase determination in protein crystallography. Krypton is not as heavy as xenon, but its K-absorption edge is situated at a wavelength (0.86 Angstrom) that is readily accessible on synchrotron radiation sources. As a test case, X-ray diffraction data at the high-energy side of the K edge were collected on a crystal of porcine pancreatic elastase (molecular weight of 25.9 kDa) put under a krypton gas pressure of 56 x 10(5) Pa. The occupancy of the single Kr atom is approximately 0.5, giving isomorphous and anomalous scattering strengths of 15.2 and 1.9 e, respectively. This derivative could be used successfully for phase determination with the SIRAS method (single isomorphous replacement with anomalous scattering). After phase improvement by solvent flattening, the resulting electron-density map is of exceptionally high quality, and has a correlation coefficient of 0.85 with a map calculated from the refined native structure. Careful data collection and processing, as well as the correct statistical treatment of isomorphous and anomalous signals have proven to be crucial in the determination of this electron-density map. Heavy-atom refinement and phasing were carried out with the program SHARP, which is a fully fledged implementation of the maximum-likelihood theory for heavy-atom refinement [Bricogne (1991). Crystallographic Computing 5, edited by D. Moras, A. D. Podjarny & J. C. Thierry, pp. 257-297. Oxford: Clarendon Press]. It is concluded that the use of xenon and krypton derivatives, when they can be obtained, associated with statistical heavy-atom refinement will allow one to overcome the two major limitations of the isomorphous replacement method i.e. non-isomorphism and the problem of optimal estimation of heavy-atom parameters.