Protein folding is a very complex process. Even for proteins that populate only two-states-folded and unfolded - there are many pathways that connct the conformational ensembles of the two states. To understand the mechanism of protein folding, therefore, it is necessary to characterize the individual pathways, called transition paths. The transition path corresponds to the rare molecular trajectory that crosses a barrier between the two states in a which is a unique single molecule property and invisible to ensemble measurements, has never been observed experimentally for any molecular system in the condensed phase, due to the insufficient time resolution of typical single molecule experiments. As a first step toward observing transition paths, I have used single molecule Förster resonance energy transfer (FRET) spectroscopy to determine average transition path times. By developing a statistical method based on the photon-by-photon maximum likelihood analysis and carrying out a collective analysis of a large number of transitions, it was possible to dramatically improved the time resolution and determined a transition-path time of 2 μs for the WW domain (fast folding protein) and an upper bound of 10 μs for protein GB1 (slow folding protein). Surprisingly, the transition-path times for the two proteins differ by less than 5-fold while the folding rates differ by a factor of 10,000. This result shows that a slow-floding protein can fold almost as fast as a fast-folding protein when folding actually occurs. On the other hand, the transition path time of an all-α helical protein α₃D (12 μs) is much longer than those of the other two protins. The extremely low solvent viscosity (ŋ)-dependence of the transition path time(t_TP ∝ ŋ^0.3 ) suggests that this longer transition path time results from slow diffusion during barrier crossing due to increasde internal friction ("rougher" energy landscape). Lowering the pH to neutralize 12 carboxylates eliminates potential salt bridges and reduces the transition path time and the viscosity dependence to that previously observed (∝ ŋ^0.6 ) for other all-α helical proteins. These results provie the first glimpse of the structural origin for internal friction in protein foldin, suggesting that the slow barrier crossing of α₃D arises form making and breaking non-native salt-bridges during the transition path, as observed in MD simulations by the Shaw group. Future expriments using multi-color FRET will resolve individual transition paths and their distribution.