1. The relationship between the rate of evoked transmitter release and the extra-cellular concentration of Ca ions, [Ca2+]o, was studied at surface neuromuscular junctions of the frog cutaneous pectoris muscle. The average quantal content of the end-plate potential was reduced to low levels by reducing [Ca2+]o and adding 2 mM-Mn2+, 45 MM-Co2+ or 10 mM-Mg2+. 2. When the motor nerve was stimulated at a low frequency (0.5--2 Hz) in 2 mM-Mn2+ or 4 mM-Co2+, the average quantal content of evoked release was proportional to the fourth power of [Ca2+]o down to the lowest measurable quantal contents, around 2--4 quanta per 1000 stimuli. Combined with previous studies, this result indicates that evoked transmitter release has a steep, nonlinear dependence on [Ca2+]0 over our orders of magnitude of evoked release. 3. Calculations predict that if evoked and spontaneous release have the same fourth power dependence on intracellular [Ca2+], then the curve relating evoked release and [Ca2+]o should become much less steep as the evoked release rate approaches the spontaneous release rate. Our observation that the relationship between evoked release and [Ca2+]o remains fourth power down to very low release rates suggests that most spontaneous quantal release does not have the same dependence on intracellular [Ca2+], or does not use the same intracellular Ca2 pool, as evoked release. 4. In 2--10 mM-Mg2+, the lowest average quantal contents were markedly higher than the fourth power prediction. This discrepancy may occur either because Mg2+ somehow elevates intracellular [Ca2+], or because Mg2+ is itself a weak activator of transmitter release. 5. Even at very low rates of evoked release, increasing the stimulus frequency to 5--50 Hz caused a progressive increase in both evoked release and the rate of 'background' quantal release during the interstimulus interval. The frequency-dependent enhancement of both evoked and background release was more pronounced in solutions containing 10 mM-Mg2+ than in solutions containing 2 mM-Mn2+.