1. Transmitter release at the neuromuscular junction of the crayfish walking leg has been found to deviate from binomial predictions immediately after the onset of repetitive stimulation of the presynaptic nerve at frequencies of at least 15 Hz. 2. After several minutes of continuous stimulation, and at lower rates of stimulation, however, the number of quanta released could be described quite well by binomial or Poisson statistics. 3. Deviations from the theoretical expectations were characterized by (a) fewer release failures than predicted, (b) occasions in which the number of quanta released was more than the estimated number of quanta available for release, and (c) a tendency for the data to be underdispersed. 4. Each of these three characteristics are consistent with the hypothesis that different releasable quanta may have different probabilities of responding to a nerve impulse. 5. Using two different methods, different values of the non-uniform probabilities were estimated from the data. At each synaptic site at least one of the estimated probabilities was very high. 6. The need for caution in interpreting statistical description of quantal release is emphasized.