Circular holes in industrial parts are a very important feature which are often used as assembly references for other parts, so accurate measurement is required. Measurement of circular holes based on computer vision is a promising method but also very challenging, because it is tricky to find the edge of a hole accurately in an image. In this paper, an accurate and robust method for the measurement of circular holes is proposed. Firstly, highly trusted edge points of the hole are found based on pixel intensity information. Secondly, a new energy function model based on the epipolar geometry constraint is introduced to obtain the optimal edges of the hole. Finally, a spatial circular hole is accurately reconstructed by using the points on the optimized edges as input to a three-dimensional reconstruction method. Our experimental result shows that the proposed method could measure various circular holes from different views with an error of around 0.05?mm.