Coal transport properties are highly dependent on the underlying fractured network, known as cleats, which are characterized by geometrical and topological properties. X‐ray microcomputed tomography (micro‐CT) has been widely applied to obtain 3‐D digital representations of the cleat network. However, segmentation of 3‐D data is often problematic due to image noise, which will result in inaccurate estimation of coal properties (e.g., porosity and specific surface area). To circumvent this issue, a discrete fracture network (DFN) model is proposed. We develop a characterization framework to determine if the developed DFN models can preserve the topological properties of the coal cleat network found in micro‐CT data. We compute the Euler characteristic, fractal dimension, and percolation quantities to analyze the topology locally and globally and compare the results between micro‐CT data (before denoising), filtered micro‐CT data (after denoising), and the DFN model. We find that micro‐CT data with noise have extensive connectivity while filtered micro‐CT data and DFN models have similar topology both globally and locally. It is concluded that the topology of the DFN models are closer to that of the realistic cleat network that do not have segmentation‐induced pores. In addition, micro‐CT imaging always struggles with the trade‐off between sample size and resolution, while the presented DFN models are not restricted by imaging resolution and thus can be constructed with extended domain size. Overall, the presented DFN model is a reliable alternative with realistic cleat topology, extended domain size and favorable data format for direct numerical simulations.