We present a novel type of spectral diffusion experiment in the millikelvin range to characterize the energy landscape of a protein as compared with that of a glass. We measure the time evolution of spectral holes for more than 300 hr after well-defined initial nonequilibrium conditions. We show that the model of noninteracting two-level systems can describe spectral diffusion in the glass, but fails for the protein. Our results further demonstrate that randomness in the energy landscape of a protein shows features of organization. There are ''deep minimum'' states separated by barriers, the heights of which we are able to estimate. The energy landscape of a glass is featureless by comparison.The folding routes, the three-dimensional structure, the dynamics, and the specific functions of a protein are all determined by the protein's energy landscape. This landscape is very complex and, hence, very difficult to measure (1), even qualitatively. What is well known, so far, is that despite the high level of structural organization (2), there is randomness in the energy landscape (3), which is reflected, for instance, in a variation of Debye-Waller factors in x-ray diffraction (4, 5), in inhomogeneous line broadening (6-8), in dispersive kinetics (9-11), and in spectral diffusion (6,8,(12)(13)(14). Much less is known about whether the randomness itself shows features of organization. It was suggested over a decade ago that the energy landscape of a protein may be organized in a hierarchical fashion (15). Recently, we found evidence for such self-similar features, although only in a qualitative sense, based on spectral diffusion experiments in the millikelvin range (13). The results showed that features in the energy landscape that were found at temperatures where the molecules become physiologically active (16) were also found in experiments below 1 K. These features are the roughness of the energy surface and the so-called deep minimum states. Quite recently, stimulated echo experiments on Zn-protophorphyrin-IXsubstituted myoglobin have shown that the self-similar organization of the energy landscape may even hold in a quantitative sense (17,18).