A model of the secretion of a quantum at a release site is proposed in which, following the influx of calcium ions, synaptic vesicles are made available for release by the activation of k phosphorylation steps with rate $\alpha $. At any time during this process the vesicles may become unavailable for secretion at rate $\gamma $. On completion of the k phosphorylation steps the vesicles participate in the formation of a fusion pore with the terminal membrane to give exocytosis at rate $\delta $. Changes in $\alpha $, $\delta $ and k are shown to produce characteristic changes in the number and timecourse of quantal secretions following a nerve impulse, which are similar to those observed following drug treatments that are thought to act selectively on each of these processes. The number of quanta secreted from nerve terminals that consist of many release sites does not fluctuate much during a low frequency train of impulses: the variance is small compared with the mean level, so secretion follows binomial rather than Poisson statistics. A theory is derived that shows that variations in the probability of secretion amongst these release sites of any particular kind fails to reduce the variance of the total secretion from the terminal; Poisson rather than binomial statistics then still apply. The theory shows that an interaction between release sites is required to reduce this variance and such an effect is provided if secretion at a site inhibits secretion at nearby sites. Simulations show that incoporating this process of autoinhibition into the model reproduces the experimental observations on the effects of calcium ions on the binomial parameters p and n as well as on the relative constancy of p during facilitation and depression of quantal secretion. Methods for estimating the timecourse of changes in the probability of secretion at release sites following an impulse, by using either the time of occurrence of first, second, third or later quantal latencies, are given. These procedures show that current methods for estimating the time-dependent probability changes are inadequate for detecting interaction between release sites, such as autoinhibition, unless this is relatively large. Therefore, estimates from third quantal latencies are used.