A method for calculating quantal size based on spectral analysis of postsynaptic potential (PSP) amplitude distributions was developed and tested by Monte-Carlo techniques. It was combined with a test to reveal the bias of the estimate of quantal size and to determine whether the peaks in amplitude distributions reflect quantal release or a sampling artifact. Spectral density was computed using fast Fourier transformation after subtraction of a fitted polynomial from the probability density function. The method overestimated quantal size for values less than two standard deviations of noise, indicating that those estimates as well as estimates of quantal size computed for examples of non-quantal distributions are not reliable. This was the case for 34 of 36 sets of sensorimotor excitatory PSPs of the frog, suggesting that most values of the quanta in synapses between primary fibres and lumbar motoneurons are smaller than 70-90 microV.