In this paper, an accurate image processing algorithm for determination of volume and surface area of orange is developed. The proposed machine vision system consists of two CCD cameras, an appropriate lighting system and a personal computer. The cameras are placed at right angle to each other in order to give two perpendicular views of the image of the orange. Initially, the algorithm segments the background and divides the image into a number of frustums of right elliptical cone. The volume and surface area of each frustum are then computed by the segmentation method. The total volume and surface area of the orange is approximated as the sum of all elementary frustums. The difference between the computed volumes and surface areas obtained by the image processing method and measured by water displacement and tape method, respectively, are not statistically significant at the 5% level. The Bland-Altman results show that the orange size has no effect on the accuracy of estimated volume and surface area found by the image processing technique. The regression formula, M=0.68VIP+44.6, between the computed volume and the measured mass of oranges is found to be highly correlated with R2=0.93.