The primary grain size is an important parameter with respect to the internal quality of steel. The examination of the primary grain size by hand is very time consuming and also a subjective process. Therefore, a digital image processing software has been developed using MATLAB R2007b. Existing machine vision based methods are just able to evaluate secondary or tertiary grain boundaries because of their visible shape (i.e. pro-eutectoid ferrite). However, these methods are not able to detect primary grain boundaries. The software developed during this Master Thesis is, to the best of my knowledge, the first automatic procedure for the determination of primary grain boundaries from the orientation of the dendrites.