Light inspection machine is a detecting instrument for visible impurities in liquid by machine vision technology. In order to establish a calibration curve between the particle sizes of impurities in the liquid and their image sizes in the intelligent inspection machine, a calibration method of particle sizes was proposed based on orthogonal binocular vision. First, the correspondence relationship of the impurities in the orthographic camera images was created based on the image spatial location matching and 3D spatial coordinates of the impurity particles in images were obtained. Then, according to the obtained three-dimensional coordinates of the impurity particles in liquid images, one-dimensional magnification errors of cylindrical liquid were calculated, the geometric correction of impurity particle images was completed and the defocused error coefficient was also obtained. Furthermore, the precise values of image sizes of the impurity particles were calculated. Finally, the standard particles with a diameter of 50 μm in a penicillin bottle were calibrated. Calbiration results show that the dispersion coefficients of impurity particle sizes are reduced from 0.11 to 0.05. The experiments indicate that this method can get a particle size calibration curve with a high precision and can help the light inspection machine to measure the particle sizes of impurities in the liquid accurately.