1. EPSCs were recorded under whole-cell voltage clamp at room temperature from Purkinje cells in slices of cerebellum from 12- to 14-day-old rats. EPSCs from individual climbing fibre (CF) inputs were identified on the basis of their large size, paired-pulse depression and all-or-none appearance in response to a graded stimulus. 2. Synaptic transmission was investigated over a wide range of experimentally imposed release probabilities by analysing fluctuations in the peak of the EPSC. Release probability was manipulated by altering the extracellular [Ca2+] and [Mg2+]. Quantal parameters were estimated from plots of coefficient of variation (CV) or variance against mean conductance by fitting a multinomial model that incorporated both spatial variation in quantal size and non-uniform release probability. This 'multiple-probability fluctuation' (MPF) analysis gave an estimate of 510 +/- 50 for the number of functional release sites (N) and a quantal size (q) of 0.5 +/- 0.03 nS (n = 6). 3. Control experiments, and simulations examining the effects of non-uniform release probability, indicate that MPF analysis provides a reliable estimate of quantal parameters. Direct measurement of quantal amplitudes in the presence of 5 mM Sr2+, which gave asynchronous release, yielded distributions with a mean quantal size of 0.55 +/- 0.01 nS and a CV of 0.37 +/- 0.01 (n = 4). Similar estimates of q were obtained in 2 mM Ca2+ when release probability was lowered with the calcium channel blocker Cd2+. The non-NMDA receptor antagonist 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX; 1 microM) reduced both the evoked current and the quantal size (estimated with MPF analysis) to a similar degree, but did not affect the estimate of N. 4. We used MPF analysis to identify those quantal parameters that change during frequency-dependent depression at climbing fibre-Purkinje cell synaptic connections. At low stimulation frequencies, the mean release probability (pr) was unusually high (0.90 +/- 0.03 at 0.033 Hz, n = 5), but as the frequency of stimulation was increased, pr fell dramatically (0.02 +/- 0.01 at 10 Hz, n = 4) with no apparent change in either q or N. This indicates that the observed 50-fold depression in EPSC amplitude is presynaptic in origin. 5. Presynaptic frequency-dependent depression was investigated with double-pulse and multiple-pulse protocols. EPSC recovery, following simultaneous release at practically all sites, was slow, being well fitted by the sum of two exponential functions (time constants of 0.35 +/- 0.09 and 3.2 +/- 0.4 s, n = 5). EPSC recovery following sustained stimulation was even slower. We propose that presynaptic depression at CF synapses reflects a slow recovery of release probability following release of each quantum of transmitter. 6. The large number of functional release sites, relatively large quantal size, and unusual dynamics of transmitter release at the CF synapse appear specialized to ensure highly reliable olivocerebellar transmission at low frequencies but to limit transmission at higher frequencies.