In this paper protein crystallization is regarded as a process starting with phase separation in a two-component system. The nucleation time of a lysozyme solution is measured by recording the NMR spectra of crystallizing solutions as a function of time. It is found that there is an appreciable induction time before the first nuclei appear in the solution and that this induction time depends strongly on the degree of supersaturation due to the protein concentration at a given ionic strength or due to the temperature. From the experimental data it is evident that (at least for lysozyme) crystallization under the prevailing experimental conditions is a transient process with an induction time and not a steady-state process.