The mechanism of folding of dihydrofolate reductase from Escherichia coli was reinvestigated by studying the unfolding and refolding kinetics using absorbance and fluorescence spectroscopies. The original kinetic model proposed that folding involved a series of native, intermediate, and unfolded forms which interconverted through four independent channels linked by slow cis/trans isomerization reactions at Xaa-Pro peptide bonds [Touchette, N. A., Perry, K. M., & Matthews, C. R. (1986) Biochemistry 25, 5445]. Recently, alternative sequential models have been proposed [Frieden, C. (1990) Proc. Natl. Acad. Sci. U.S.A. 87, 4413; Kuwajima et al. (1991) Biochemistry 30, 7693] which challenge the original proposal. Stopped-flow studies of the intrinsic tryptophan fluorescence demonstrated the presence of three (and tentatively four) kinetic phases in unfolding which correlated well with four phases previously observed in refolding experiments. By monitoring the binding of the inhibitor methotrexate during folding at varying relative concentrations of inhibitor to protein, it was found that the selective loss of the slow-folding phases at substoichiometric levels could only be explained by a four-channel folding model. Double-jump experiments (native-->unfolded-->native) showed that the four refolding channels are populated within 20 s at 15 degrees C and are not likely to be due to proline isomerization. Reverse double-jump experiments (unfolded-->native-->unfolded) demonstrated that interconversions between native conformers are more rapid than originally proposed. Interestingly, the majority of the protein folds through a channel to a native conformer that is minimally populated at equilibrium. This implies that although the folding of dihydrofolate reductase is ultimately under thermodynamic control, kinetic factors contribute to the transient populations of native species during folding.