CNC machine tools which use conventional feedback controllers usually result in significant tracking error when machining a contour, especially at high feed speed. This paper presents for the first time a new methodology for compensating general contour error. The new method uses contour error which is measured in advance to improve the feed drive accuracy of CNC machine tools. The eventual purpose of this study is to develop an autonomous control system for CNC machine tools, which should be able to perform precisely under a certain range of machining parameters. The contour error is considered to be related to several kinds of machining parameters and a neural network is used here, as the intelligent part of the autonomous control system, to learn the relationship between the contour error and machining parameters. This paper gives results of both simulation and experiments. © 1995, The Japan Society of Mechanical Engineers. All rights reserved.