The relation between quantal transmitter release and presynaptic Ca2+/Ba2+ entry at the mouse neuromuscular junction was studied, making use of the finding that in the presence of Ba2+ trains of nerve stimuli or brief nerve terminal depolarizations elicit "tails" of raised miniature end-plate potential frequency (fm) that reflect entry of Ba2+ per pulse, and hence effectiveness of pulses in opening Ca2+/Ba2+ channels; at the same time these pulses elicit end-plate potentials. With nerve stimulation in the presence of Ba2+ and Ca2+ and modulation of release by raised Mg2+ or bekanamycin, slopes of log quantal content (m) vs log apparent Ba2+ entry per pulse were close to 4, which is the same as the Hill coefficient for Ba2+ cooperativity derived from other data. With depolarizing pulses of varied intensity, however, similar plots gave slopes close to 2, with Ba2+ alone or in a mixture of Ca2+ and Ba2+. Thus, the relation between transmitter release and Ca2+ (or Ba2+) entry apparently depends upon how entry is varied; varying the numbers of channels opened is not the same as varying ion entry per channel. A mathematical model was developed to examine the consequences of heterogeneity of local Ca2+ (or Ba2+) between release sites, arising because of stochastic variation of number and time course of Ca2+ channels opened per site; the experimental results were consistent with this model. It was therefore concluded that release is normally governed by intracellular Ca2+ close to points of Ca2+ entry through channels; stochastic factors give rise to more release than if Ca2+ were homogeneously distributed. If Ca2+ channels are uniformly close to release sites the average number of channels opened per site per action potential may be as low as 4.