Classical and quantum statistical mechanics are cast here in the language of projective geometry to provide a unified geometrical framework for statistical physics. After reviewing the Hilbert–space formulation of classical statistical thermodynamics, we show that the specification of a canonical polarity on the real projective space RPn induces a Riemannian metric on the state space of statistical mechanics. In the case of the canonical ensemble, equilibrium thermal states are determined by a Hamiltonian gradient flow with respect to this metric. This flow is characterized by the property that it induces a projective automorphism on the state manifold. The measurement problem for thermal systems is studied by the introduction of the concept of a random state. The general methodology is then extended to establish a new framework for the quantum–mechanical dynamics of equilibrium thermal states. In this case, the relevant phase space is the complex projective space CPn, here regarded as a real manifold γ endowed with the Fubini–Study metric and a compatible symplectic structure. A distinguishing feature of quantum thermal dynamics is the inherent multiplicity of thermal trajectories in the state space, associated with the non–uniqueness of the infinite–temperature state. We are then led to formulate a geometric characterization of the standard KMS relation often considered in the context of C* algebras. Finally, we develop a theory of the quantum microcanonical and canonical ensembles, based on the geometry of the quantum phase space γ. The example of a particle of quantum spin–1/2 particle in a heat bath is studied in detail.