An evolutionary process is simulated with a simple spin-glass-like model of proteins to examine the origin of folding ability. At each generation, sequences are randomly mutated and subjected to a simulation of the folding process based on the model. According to the frequency of local configurations at the active sites, sequences are selected and passed to the next generation. After a few hundred generations, a sequence capable of folding globally into a native conformation emerges. Moreover, the selected sequence has a distinct energy minimum and an anisotropic funnel on the energy surface, which are the imperative features for fast folding of proteins. The proposed model reveals that the functional selection on the local configurations leads a sequence to fold globally into a conformation at a faster rate.Natural proteins are known to fold into unique threedimensional structures. Though this folding ability must have emerged in the course of evolution, most investigations have focused not so much on the origin as on the mechanism of the folding process.In dealing with the mechanism of the folding process, recent theoretical work with lattice models (1-4) and with spin-glass ideas (5-8) suggests that completely random heteropolymers do not have the same tendency to fold into a unique conformation as natural proteins do. To attain a single conformation from all the possible ones within a biological time scale, the peptide chain must have an energetically highly distinctive minimum and a so-called folding funnel (2)-i.e., a biased energy surface. This design principle of the energy landscape is called the principle of minimum frustration (5-7). Several hypotheses have been presented of the origin of minimally frustrated peptides (9-11).Partial successes in obtaining catalytic antibodies (12) and some functional peptides (13, 14) through functional selection imply the possibility that most polypeptides, if not all, have been tuned through positive selections on function during their evolution. Therefore, it is worth asking whether the existence of an accessible unique conformation evolves from random polypeptides when sequences are selected on the basis of their function alone. In fact, one of the authors and his colleagues (15) synthesized random peptides with 140 residues and showed that 10% of those peptides were water soluble. Some of the soluble peptides have weak hydrolyzing activity whose strength depends on the sequence of the random proteins (unpublished results).The function of a protein is, in general, governed directly by some active site residues. The spatial configuration of each active site is arranged in a certain way to possess a high functional activity. Then, one could ask whether such local configurational adaptation dictates the folding of the whole protein. In other words, does a globally properly folded peptide chain evolve through the restriction on the local configurations? Pande et al. (10) raised a related question whether sequences with a lower total energy of the protein-subst...