In order to improve the accuracy of a recovered object by machine vision system,the relationship between the camera position and the accuracy of the recovered object is analyzed and the weighted values based on the relationship are evaluated. Then, the weighted least square method is used in the machine vision system.After expressing the machine vision model by parameter equations, a recovering equation for 3D space point recovering is deduced based on the parameter equation.Then,according to the error transmission principle, the transmission from the uncertainty of a camera image plane to the uncertainty of an observe space is analyzed,and the relationship between the error transmission and the camera position is obtained.Finally,the covariance of the observed object is used to evaluate the weighted value for the weighted least square method.Experimental results indicate that the accuracy of the weighted least square method is better than that of the general least square method. When the noise variance is bigger than 0.5 and the number of the photos is less than 30, this method can offer a very good measuring accuracy.According to the experiment,when the measurement distances are 37.031 0 cm,24.970 4 cm and 26.015 3 cm,the weighted least square method can improve the accuracy by 0.4 cm.