Atomistic models for friction suffer from the severe length- and time-scale restrictions of molecular dynamics. In this paper, a novel approach to quantify the scratching work and the energy associated with plastic activity is used. The approach is combined with a statistical criterion to determine the significance of simulation box size, microstructure and sliding rate effects on the friction coefficient. This method is applied to a large parametric molecular dynamics study of single-asperity nanoscratch on monocrystalline and polycrystalline aluminium substrates. The results show that even though simulation size affects the plastic core mechanisms of sliding friction, the method overcomes size dependence when it comes to the predicted value of the friction coefficient. We show that friction in monocrystalline and polycrystalline substrates activates very different mechanisms and describe them in some detail for the polycrystalline case. Furthermore, we show that even when the friction coefficient appears to be size independent over a range of simulation box sizes, the plastic activity associated with it remains non-monotonically size dependent. (C) 2013 Elsevier Ltd. All rights reserved.