This paper describes an experimental investigation into the patterns of groove wear development and their monitoring at the minor cutting edge of metal-cutting tools. It is shown that these grooves, once formed, adversely affect the dimensional accuracy and surface roughness in finish machining. Multivariate time series stochastic models are developed directly from vibration signals, measured with a miniature three-dimensional (3D) accelerometer mounted in close vicinity to the tool tip. Based on the established models, multiple dispersion analysis is introduced to isolate ingredients sensitive to particular stages of groove wear from the 3D vibrations, and to quantify their relations. The results are physically interpreted and show that the method presented in this paper is a feasible means for on-line monitoring of groove wear development in automated finish machining. © 1991.