In the present study, Polyamide 12 was compounded with glass fibers in both laboratory and industrial twin screw extruders using various processing conditions (screw speed and feed rate). Dead-stop experiments were performed and samples were collected at different locations along the screws in order to determine the fiber length distribution and the extent of fiber breakage. Results show that significant fiber break-up occurs right after the addition of glass fibers to the molten matrix. Similarly fiber length distribution changes drastically at the first sampling location, near the glass fiber feeder. Processing conditions also influence the extent of degradation: it increases with screw speed and decreases with feed rate, which controls the residence time. Flow modelling has been used to calculate the flow conditions along the screw profile. It is shown that the modified Shon-Liu-White model previously proposed to describe the average fiber length evolution as function of specific energy is not able to correctly predict the evolution along the screw profile. In similar processing conditions, the large industrial extruder appears as less severe than the small laboratory one.