Understanding the mechanism of protein folding is often referred to as the second half of genetics. Computational approaches have been instrumental in the efforts. Simplified models have been applied to understand the physical principles governing the folding processes and will continue to play important roles in the endeavor. Encouraging results have been obtained from all-atom molecular dynamics simulations of protein folding. A recent microsecond-length molecular dynamics simulation on a small protein, villin headpiece subdomain, with an explicit atomic-level representation of both protein and solvent, has marked the beginning of direct and realistic simulations of the folding processes. With growing computer power and increasingly accurate representations together with the advancement of experimental methods, such approaches will help us to achieve a detailed understanding of protein folding mechanisms.