The objective of this study is to evaluate the capability of two modelling approaches in capturing both accelerations and deformations from head impacts, and to evaluate the effect of modelling approach on brain injury prediction. The first approach is a so‐called smeared technique, in which the properties of the two glass sheets and the intermediate polyvinyl butyral (PVB) are combined and divided into two coinciding shell layers, of which one can fracture. The second approach consists of three shell layers, representing the glass and PVB, separated by the distance of their thickness, using a non‐local failure criterion to initiate fracture in the glass. The two modelling approaches are compared to impact experiments of flat circular windscreens, measuring deformations and accelerations as well as accelerations from impacts against full vehicle windscreens. They are also used to study head‐to‐windscreen impacts using a detailed Finite Element (FE) model, varying velocity, impact direction and impact point. Only the non‐local failure model is able to adequately capture both the accelerations and deformations of an impactor. The FE head model simulations also reveal that the choice of modelling approach has a large effect on the both localisation of the strain in the brain and the characteristics of the strain‐time curve, with a difference in peak strain between 8% and 40%.