The sieving by vision method was used, in a first time, to measure the mean of the real cutting length of maize silage particles. This vision method was studied in detail, working with calibrated stone images, in order to analyse the quality of this method to extract from images different information, concerning the composition, or the granulometry on an image. Finally, we saw that this method permits to obtain three kind of information on a maize silage image: 1) Mean size of the silage particles obtained from the modelled curve for a given sieve size. This sieve size can be 13mm. 2) Heterogeneity coefficient of the image, by considering the distribution curve for different sieve sizes (for example 7, 10, 13 and 16mm) for this image and the distribution curve corresponding to the best as possible as the mean size of the maize silage particles on the image, found in 1). We make an error calculation, for the different sieve sizes between both curves, to obtain this coefficient. 3) Estimation of presence rates for small (for example under 7mm), medium (for example between 7 and 16mm) and big (for example over 16mm) particles on an image, using the results obtained for two sieve sizes (for example 7mm and 16mm).