We investigate the kinetic pathways of folding and unfolding of the designed miniprotein Trp-cage in explicit solvent. Straightforward molecular dynamics and replica exchange methods both have severe convergence problems, whereas transition path sampling allows us to sample unbiased dynamical pathways between folded and unfolded states and leads to deeper understanding of the mechanisms of (un)folding. In contrast to previous predictions employing an implicit solvent, we find that Trp-cage folds primarily (80% of the paths) via a pathway forming the tertiary contacts and the salt bridge, before helix formation. The remaining 20% of the paths occur in the opposite order, by first forming the helix. The transition states of the rate-limiting steps are solvated native-like structures. Water expulsion is found to be the last step upon folding for each route. Committor analysis suggests that the dynamics of the solvent is not part of the reaction coordinate. Nevertheless, during the transition, specific water molecules are strongly bound and can play a structural role in the folding.protein folding ͉ reaction coordinate ͉ transition path sampling ͉ replica exchange ͉ transition state ensemble E lucidating the mechanism by which proteins fold into their native state remains a central issue in molecular biology. For single domain two-state folding proteins, several decades of experimental, theoretical, and simulation studies have revealed two major qualitative folding mechanisms. In the diffusion-collision mechanism (1), proteins first form secondary structure elements followed by a diffusive search toward the tertiary native state structure. In the nucleation-condensation mechanism (2), a nucleus of crucial tertiary contacts is made, around which the native structure condensates. In recent years, these two mechanisms were combined in a unified view (3).By bridging the gap between experiments and computer simulation, the discovery of small and fast folding proteins has contributed much to the understanding of generic folding mechanisms. The fastest of those is the designed 20-residue miniprotein Trp-cage (NLYIQ WLKDG GPSSG RPPPS) (4), which folds in 4 s to a native state with an ␣-helix, a salt bridge, and a polyproline II helix shielding the central tryptophan from solvent. Laser temperaturejump spectroscopy experiments by Qiu et al. (5) indicated two-state folding. Subsequently, fluorescent correlation spectroscopy by Neuweiler et al. (6) revealed that the protein (un)folds in a more complicated manner via an intermediate molten globule-like state, characterized by exposure of the tryptophan to the solvent. It remains unclear at what stage of folding the helix is being formed. Recent UV-resonance Raman spectroscopy measurements show some evidence of a helical structure in the denaturated state of Trp-cage, and thus suggest that an early formation of the helix is possible (7). Many molecular dynamics (MD) simulations were performed to investigate thermodynamic stability of the protein and elucidate possible folding p...