In unmanned machining, when an abnormality condition occurs in a tooling system, a proper action to detect, interpret and determine the condition has to be made. In this paper, a method of sensing the abnormal conditions of tools on the basis of image processing is developed. To recognize the tool conditions, a pattern recognition technique using the multi-layered perceptron with the back-propagation algorithm as a neural network is applied. This paper also presents quantitative measurement of flank wear and crater wear in order to use the data as the input of the neural network. (C) 1998 Elsevier Science Ltd. All rights reserved.