Recently available G-protein coupled receptor (GPCR) structures and biophysical studies suggest that the difference between the effects of various agonists and antagonists cannot be explained by single structures alone, but rather that the conformational ensembles of the proteins need to be considered. Here we use an elastic network model-guided molecular dynamics simulation protocol to generate an ensemble of conformers of a prototypical GPCR, β2-adrenergic receptor (β2AR). The resulting conformers are clustered into groups based on the conformations of the ligand binding site, and distinct conformers from each group are assessed for their binding to known agonists of β2AR. We show that the select ligands bind preferentially to different predicted conformers of β2AR, and identify a role of β2AR extracellular region as an allosteric binding site for larger drugs such as salmeterol. Thus, drugs and ligands can be used as "computational probes" to systematically identify protein conformers with likely biological significance.