A machine vision system is proposed to measure the diameter of a mini milling cutter on line, and the measurement uncertainty is analyzed in this paper. The system consists of a back-light source, a rotate stage, a rotary encoder, a camera and corresponding software. While the milling cutter rotates, the camera is triggered by the rotary encoder and a series of images are captured. The sources of uncertainty and their effects are analyzed, including the parallelism of the light source, lens distortion, CCD noise, focusing situation, the geometric properties of tested objects and the algorithm design. According to results of experiments, the uncertainty mainly depends on the geometric property of the tested object when sampling rate is not enough, otherwise, it is mainly relative to the parallelism of the light source. In the experiment performed, the system uncertainty is 4 μm with a sampling rate of 33 frame/circle, both accuracy and efficiency meet the requirement of manufacturing.