Chronically implanted microelectrodes have been used to obtain extracellular records of trains of spontaneous action potentials from 30 neurones in the cerebral cortices of 13 unrestrained cats. Recorded neurones were in or near to primary visual cortex, primary auditory cortex, and in the supra-sylvian gyrus. Records were made with animals in several different behavioural states, which included sleep with rapid eye-movements and quiet sleep, peacefully awake, and alarmed. Interval distributions derived from trains of 200 action potentials recorded in less than 80 s did not differ significantly from curves in which the probability of any interval is normally distributed about a modal interval, when plotted on a logarithmic time-axis. Thus the complete interval distributions of neurones firing faster than 2.5/s can be described by two parameters - a modal interval, and a geometric standard deviation. This quantitative description of interval distributions proved equally applicable to neurones in all three cortical areas and was valid over the whole range of behavioural states examined. It does not usually hold when the discharge frequency of a neurone is lower than 2.5 action potentials per second. An acceptable fit for a log-normal curve can then only be obtained for intervals that are shorter than about ten times the modal interval. It is pointed out that mean frequency of discharge is a measure of neural activity which is a secondary parameter, since it is dependent upon both modal interval and geometric standard deviation. Our preliminary data show that the two parameters which define the best-fit log-normal curves can vary independently with the behavioural state of the animal in a way that suggests that they may be physiologically important.