Ball mills have a low efficiency rate partially due to the lack of a proper method to monitor the mill filling level, which makes it difficult to control the grinding process. The purpose of this paper is to explore the possibilities that DEM offers to characterize the load torque of tumbling ball mills in the frequency domain and, from this characterization, to establish the basis of a methodology capable of estimating the mill filling level. To achieve this, a pilot scale ball mill was considered and a campaign of simulations and experimental tests was carried out in dry grinding conditions, considering variables such as the rotation speed of the mill, the type of particle to be grinded, and the mill filling level. For each simulation, spectral analysis of torque data generated by DEM software was performed. The results obtained from simulations and subsequent torque data processing show the possibility of characterizing the load torque of ball mills in the frequency domain without the influence of components alien to the grinding process. In this paper, it is shown that load torque signal in ball mills contains enough information to characterize, unambiguously, the load level of the mill. Thus, a methodology to evaluate the mill filling level based on torque spectral analysis is proposed. (C) 2017 The Author(s). Published by Elsevier B.V.