Researchers and engineers have widely adopted the discrete element method (DEM) for simulation of bulk materials. One important aspect in such simulations is the determination of suitable material and contact law parameters. Very often, these parameters have to be calibrated because they are difficult to measure or, like rolling friction, do not have a physical analogue. Moreover, coarse-grained particle models are commonly used to reduce computational cost and these always require calibration. Despite its disadvantages, trial and error remains the usual way to calibrate such parameters. The main aim of this work is to describe and demonstrate a methodical calibration approach which is based on Latin hypercube sampling and Kriging. The angle of repose and bulk density are calibrated for spherical glass beads. One unique feature of this method is the inclusion of the simulation time-step in the calibration procedure to obtain computationally efficient parameter sets. The results show precise calibration outcomes and demonstrate the existence of a solution space within which different parameter combinations lead to similar results. Kriging meta-models showed excellent correlation with the underlying DEM model responses. No correlation was found between static and rolling friction coefficients, although this has sometimes been assumed in published research. Incorporating the Rayleigh time-step in the calibration method yielded significantly increased time-step sizes while retaining the quality of the calibration outcome. The results indicate that at least particle density, Young's modulus and both rolling and static friction coefficients should be used for calibration; trial-and-error would be highly inefficient for this number of parameters which highlights the need for systematic and automatized calibration methods.