The fact that a large data set can be summarised with just two or three parameters is useful in itself – for instance, if we want to characterise clinical observations (examples of this are given in ). But it is obviously not a factor that will drive evolutionary change. So why do reciprocals of reaction times have a normal distribution? What sort of mechanism could account for a recinormal distribution of latency? Does it make any functional sense? The model that is about to be presented was first conceived in relation to saccadic latencies (Carpenter 1981, 1988), but appears to have a much wider applicability. It will now be described briefly and then considered in more detail.