We studied the mechanism of the lamellar-to-inverted hexagonal (L alpha/H[II]) phase transition, using time-resolved cryotransmission electron microscopy (TRC-TEM), 31P-NMR, and differential scanning calorimetry. The transition was initiated in dispersions of large unilamellar vesicles of dipalmitoleoyl phosphatidylethanolamine (DiPoPE). We present evidence that the transition proceeds in three steps. First, many small connections form between apposed membranes. Second, the connections aggregate within the planes of the bilayers, forming arrays with hexagonal order in some projections. Third, these quasihexagonal structures elongate into small domains of H(II) phase, acquiring lipid molecules by diffusion from contiguous bilayers. A previously proposed membrane fusion mechanism rationalizes these results. The modified stalk theory predicts that the L alpha/H(II) phase transition involves some of the same intermediate structures as membrane fusion. The small interbilayer connections observed via TRC-TEM are compatible with the structure of a critical intermediate in the modified stalk mechanism: the trans monolayer contact (TMC). The theory predicts that 1) TMCs should form starting at tens of degrees below TH; 2) when TMCs become sufficiently numerous, they should aggregate into transient arrays like the quasihexagonal arrays observed here by TRC-TEM; and 3) these quasihexagonal arrays can then elongate directly into H(II) phase domains. These predictions rationalize the principal features of our data, which are incompatible with the other transition mechanisms proposed to date. Thus these results support the modified stalk mechanism for both membrane fusion and the L alpha/H(II) phase transition. We also discuss some implications of the modified stalk theory for fusion in protein-containing systems. Specifically, we point out that recent data on the effects of hydrophobic peptides and viral fusion peptides on lipid phase behavior are consistent with an effect of the peptides on TMC stability.