A parallel decision making mechanism that superficially resembles the sort of processing that may be characteristic of neurones in the brain. it is a pattern recognition device with a threshold. If the linear combination of the ‘weighted inputs’ is greater than some threshold value then the perceptron ‘fires’. It is possible for a perceptron to learn, eg, if the weights associated with the inputs that were active in the case of a false alarm are decreased, and weights associated with inputs that were active in the case of a miss are increased then it is intuitively plausible that recognition performance will improve; and there is a theorem that says that a perceptron will learn to recognise a class correctly over a finite number of errors. The analysis of the mathematical properties of perceptrons revealed profound limitations to their competence. These limitations are largely due to the difficulties inherent in making global decisions on the basis of only local evidence. Thus a perceptron can’t tell whether a figure is connected or not, or whether there is one. and only one. instance of a pattern present. Nevertheless there is a resurgence of interest in perceptrons associated with the development of connectionist schemes for visual processing.