Protein folding is a cooperative phenomenon where at the midtransition temperature of thermal denaturation only fully folded and fully unfolded species coexist in appreciable amounts, while the concentration of partially folded forms is negligible (1) (see Fig. 1). The implication of this finding from early calorimetric studies for kinetics was immediate and dramatic; it suggested that protein folding is a barrier-crossing process whereby folding proteins spend most of the time fluctuating idly in the unfolded state waiting for a rare fluctuation that rapidly carries it to the top of the barrier with a subsequent “downhill” run to the native state. (Fig. 1). Such a scenario was indeed discovered in folding simulations of simplified, yet nontrivial, sequence-based models in the 1990s (2). The trajectories carrying the protein over the barrier are called transition paths. It is the transition-path trajectories where all folding action occurs, and discerning them step by step holds a key to complete understanding of the mysteries of folding dynamics. Not surprisingly, many theoretical studies focused on detailed retracing of transition-path trajectories in search of mechanistic insights into protein folding dynamics (2–5). However, direct experimental observations of transition paths have been elusive because, in the ensemble, each molecule crosses the barrier at its own stochastic point in time so that an averaged picture of exponential kinetics emerges. Transition paths can be observed only in single-molecule experiments, and the work of Eaton and coworkers in this issue of PNAS (6) reports such observations for the first time. The very nature of transition paths, where folding and unfolding actually happen, makes their study a challenge for single-molecule studies. The experiments must have very high temporal resolution to resolve the very fast (and dramatic) events of reaching the top of the barrier with subsequent rapid descent to the native state. Although there have been interesting single-molecule optical studies of protein folding since the pioneering work of DeGrado, Hochstrasser, and coworkers (7) [reviewed by Schuler and Eaton (8)], the time resolution of single-molecule FRET experiments has been insufficient for observing transition paths. There are many technical difficulties related to photophysics in FRET experiments and the need to immobilize protein molecules without introducing artifacts. The fundamental issue is a peculiar “uncertainty principle” of single-molecule FRET experiments: to reach higher temporal resolution the laser intensity must be as high as possible (to increase photon count-making time bins as small as possible), but the higher laser intensity also results in faster photobleaching of the FRET chromophores that restricts the analysis to shorter trajectories that are less likely to contain transition paths, therefore providing a smaller number of interesting trajectories available for statistical analysis.