The mechanism of membrane fusion was studied by using human erythrocyte ghosts held in close contact by alternating current-induced dielectrophoresis and inducing fusion with a single electric field pulse. Individual fusion events were followed visually using either 1,1'-dihexadecyl-3,3,3',3'-tetramethylindo carbocyanine perchlorate as a membrane-mixing label or 10-kD fluorescein isothiocyanate-dextran as a contents-mixing label. However, over a range of variables, the number of contents-mixing events usually considerably exceeded the number of membrane-mixing events, although the discrepancy was less at higher ionic strength. However, when the dielectrophoretic force holding the membranes in contact was turned off after the pulse, Brownian motion caused some of the groups of ghosts in which contents mixing occurred to eventually separate from one another, showing that they could not represent fusion events. Separate experiments showed, conversely, that fusion did occur in the groups that did not separate after the dielectrophoresis was turned off.