1. We have performed a detailed quantal analysis of excitatory postsynaptic potentials (EPSPs) evoked by minimal extracellular stimulation in the CA1 region of slices of adult rat hippocampus maintained in vitro. 2. EPSPs were evoked at 2-5 Hz, and the eight that were analysed all showed at least a 50% depression of mean peak amplitude during recording. 3. EPSP amplitude fluctuations were analysed by three methods: the use of amplitude frequency histograms with clear and reliable peaks where available, graphs of the EPSP (coefficient of variation)-2 against EPSP mean, and analysis of EPSP mean and standard deviation assuming simple binomial statistics with the number of release sites (N) kept constant but the quantal size (Q) and the release probability (Pr) allowed to vary over time. 4. The results of the three analysis procedures were in good agreement. Seven EPSPs showed a substantial reduction in the mean number of quanta released per trial, and in three cases this was the predominant mechanism of the depression. Five EPSPs showed a substantial decrease in Q. Values for N ranged between 3 and 18, with a median of 6; Pr ranged between 0.14 and 0.81 and Q between 66 and 275 microV. 5. We used the Q estimates from the binomial method to correct the recorded EPSP amplitudes for changes in quantal size over time. For seven out of the eight EPSPs, this rescaling procedure allowed histograms with clear peaks to be obtained from longer runs of data, or improved the sharpness of the peaks in histograms from all the recorded data. The improvement in peak sharpness was assessed using an autocorrelation-based method. The correction was much less successful if the Q estimates were obtained with a variant of the binomial method in which Pr was held constant and N was allowed to vary. 6. The only simple explanation for the success of the correction procedure is that changes in quantal size were a major factor in obscuring peaks in histograms based on large numbers of trials, and that the quantal size estimates from the binomial method with N held constant were reasonably accurate. 7. We conclude that transmission at these synapses was quantal with relatively low quantal variance, but repetitive stimulation often induced substantial changes in the quantal parameters that might prevent the success of conventional quantal analysis approaches.