This paper presents the results of three new approaches to the problem of f
inding the folding nucleus in the given 3D protein structure. The first, si
mple phenomenological approach is based on involvement of various protein c
hain residues in the protein core and secondary structure. It is shown that
this approach demonstrates rather modest correlation with experiment. Two
other, more complicated, but physically more grounded approaches are based
on the search for the saddle points of the free energy landscape on the net
work of protein folding/unfolding pathways. To this end, one of these appro
aches employs the branch-and-bound technique, and the other makes use of dy
namic programming. These approaches, and especially the latter, give better
correlation with experiment, and the estimated free energies of the saddle
points (i.e., of the folding/unfolding nuclei) are consistent with the exp
erimentally observed rapid (within a fraction of a second) folding and unfo
lding of small proteins.