A virtual ligand‐based screening approach was designed and evaluated for the discovery of new A_2A adenosine receptor (AR) ligands. For comparison and evaluation, the procedures from a recently published virtual screening study that used the A_2A AR X‐ray crystal structure for the target‐based discovery of new A_2A ligands were largely followed. Several screening models were constructed by deriving the distinguishing structural features from selected sets of A_2A AR antagonists, so‐called frequent substructure mining. The best model in statistical terms was subsequently applied to large‐scale virtual screens of a commercial vendor library. This resulted in the selection of 36 candidates for acquisition and testing. Of the selected candidates, eight compounds significantly inhibited radioligand binding at A_2A AR (>30 %) at 10 μM, corresponding to a “hit rate” of 22 %. This hit rate is quite similar to that of the referenced target‐based virtual screening study, while both approaches yield new, non‐overlapping sets of ligands.