Lipid bilayer fusion is thought to involve formation of a local hemifusion connection, referred to as a fusion stalk. The subsequent fusion stages leading to the opening of a fusion pore remain unknown. The earliest fusion pore could represent a bilayer connection between the membranes and could be formed directly from the stalk. Alternatively, fusion pore can form in a single bilayer, referred to as hemifusion diaphragm (HD), generated by stalk expansion. To analyze the plausibility of stalk expansion, we studied the pathway of hemifusion theoretically, using a recently developed elastic model. We show that the stalk has a tendency to expand into an HD for lipids with sufficiently negative spontaneous splay, (~)J(s)< 0. For different experimentally relevant membrane configurations we find two characteristic values of the spontaneous splay. (~)J*(s) and (~)J**(s), determining HD dimension. The HD is predicted to have a finite equilibrium radius provided that the spontaneous splay is in the range (~)J**(s)< (~)J(s)<(~)J*(s), and to expand infinitely for (~)J(s)<(~)J**(s). In the case of common lipids, which do not fuse spontaneously, an HD forms only under action of an external force pulling the diaphragm rim apart. We calculate the dependence of the HD radius on this force. To address the mechanism of fusion pore formation, we analyze the distribution of the lateral tension emerging in the HD due to the establishment of lateral equilibrium between the deformed and relaxed portions of lipid monolayers. We show that this tension concentrates along the HD rim and reaches high values sufficient to rupture the bilayer and form the fusion pore. Our analysis supports the hypothesis that transition from a hemifusion to a fusion pore involves radial expansion of the stalk.