Conformational selection and induced fit are two prevailing mechanisms to explain the molecular basis for ligand-based activation of receptors. G-protein-coupled receptors are the largest class of cell surface receptors and are important drug targets. A molecular understanding of their activation mechanism is critical for drug discovery and design. However, direct evidence that addresses how agonist binding leads to the formation of an active receptor state is scarce. Here we use (19)F nuclear magnetic resonance to quantify the conformational landscape occupied by the adenosine A2A receptor (A2AR), a prototypical class A G-protein-coupled receptor. We find an ensemble of four states in equilibrium: (1) two inactive states in millisecond exchange, consistent with a formed (state S1) and a broken (state S2) salt bridge (known as 'ionic lock') between transmembrane helices 3 and 6; and (2) two active states, S3 and S3', as identified by binding of a G-protein-derived peptide. In contrast to a recent study of the β2-adrenergic receptor, the present approach allowed identification of a second active state for A2AR. Addition of inverse agonist (ZM241385) increases the population of the inactive states, while full agonists (UK432097 or NECA) stabilize the active state, S3', in a manner consistent with conformational selection. In contrast, partial agonist (LUF5834) and an allosteric modulator (HMA) exclusively increase the population of the S3 state. Thus, partial agonism is achieved here by conformational selection of a distinct active state which we predict will have compromised coupling to the G protein. Direct observation of the conformational equilibria of ligand-dependent G-protein-coupled receptor and deduction of the underlying mechanisms of receptor activation will have wide-reaching implications for our understanding of the function of G-protein-coupled receptor in health and disease.