We consider the propagation of a gravity current of density ρc at the bottom of a two-layer stratified ambient in a horizontal channel of height H, in the high-Reynolds number Boussinesq domain. The study emphasizes theoretical-analytical modeling, however, experimental and Navier-Stokes simulation data are also presented and their comparison with theory is discussed. The stratification parameters are S = (ρ1 − ρ2)/(ρc − ρ2) where ρ is the fluid density, and φ = h1R/H where h1R is the (unperturbed) ambient interface height. Here, 1 and 2 denote, respectively, the lower and upper layer and c denotes the gravity current. The reduced gravity is defined as g′ = (ρc/ρ2 − 1)g. Rigorous results are obtained for the steady-state analogue of the classical problem of Benjamin [J. Fluid Mech. 31, 209 (1968)]10.1017/S0022112068000133, in which the half-infinite gravity current has thickness h and speed U. We thereby demonstrate that the Froude number F = U/(g′h)1/2 is a function of a = h/H, S, and φ. In general, two solutions (or modes) may be realized. Issues of energy dissipation, sub- vs. supercriticality with respect to long internal waves and, more generally, the influence of upstream-propagating disturbances are discussed. For a gravity current released from a lock of height h0 and length x0, we derive an approximate shallow-water model and show that the motion is in this case governed by Ξ = H/h0, S, and φ. Although the shallow-water model neglects motion in the ambient layers and ignores the impact of propagation on stratification, the gravity current front speed in the slumping stage is in excellent agreement with measured data. Our theoretical solutions are consistent with previous results (in particular, Holyer and Huppert [J. Fluid Mech. 100, 739 (1980)] and Tan et al. [Environ. Fluid Mech. 11, 203 (2011)]), but have the advantages of being (i) derived without reliance on adjustable constants and ad hoc closures; (ii) applicable to a significantly broader range of dimensionless parameters; and (iii) better assessed by comparison against measured data. The present one-layer shallow-water approximation turns out to be a simple and versatile extension of existing models for homogeneous and linearly stratified ambients, and can be straightforwardly incorporated into the available prediction tools for gravity currents.