Neighboring ganglion cells in the retinae of vertebrates show cross-correlation between their impulse trains. Cross-correlation is found both in maintained discharges and during responses to visual stimulation. There has been speculation about the function of this statistical dependence, but little is known about its genesis. This study examines the statistics of the interimpulse intervals preceding and those following impulses that coincide with an impulse in the other train. Short intervals are rarer than expected preceding a coincidence, regardless of the form of the cross-correlation. Short intervals are more common than expected following a coincidence when the cross-correlation is positive, but rarer than expected following coincidences during negative cross-correlation. These results contradict the extant models for cross-correlation, but may be explained by the multiplicative combination of a variable common input and the variability within each cell. In addition, the lag (relative timing of coincident impulses in the two cells) is found to be related to the maintained discharge rates of the cells, implying that the lags may be explained without invoking specific delay circuits.