A combination of large-scale distributed computing efforts and Markov State Model (MSM) approaches have recently been used to simulate the folding of proteins on the millisecond timescale. These models predict new views of the folding reaction--a complex set of metastable intermediates, multiple pathways, a hub-like network structure, and compact unfolded states with residual structuring--yet, experimental observables often report simple two-state kinetics. To study the events that precede folding, we used MSM approaches to model the folding reaction of ACBP (acyl-CoA binding protein), an 86-residue helix-bundle protein that folds on the ∼10 ms timescale, with an ∼80 µs collapse phase that can be probed both by simulation and experiment. Our combined results suggest that the fast kinetic phase is not due to barrier-limited formation of a well-defined intermediate, but rather the surprisingly slow acquisition of unfolded-state structure.