Osmosis is a fundamental physical process that involves the transit of solvent molecules across a membrane separating two liquid solutions. Osmosis plays a role in many biological processes such as fluid exchange in animal cells (Cell Biochem. Biophys. 2005, 42, 277-345;1 J. Periodontol. 2007, 78, 757-7632) and water transport in plants. It is also involved in many technological applications such as drug delivery systems (Crit. Rev. Ther. Drug. 2004, 21, 477-520;3 J. Micro-Electromech. Syst. 2004, 13, 75-824) and water purification. Extensive attention has been dedicated in the past to the modeling of osmosis, starting with the classical theories of van't Hoff and Morse. These are predictive, in the sense that they do not involve adjustable parameters; however, they are directly applicable only to limited regimes of dilute solute concentrations. Extensions beyond the domains of validity of these classical theories have required recourse to fitting parameters, transitioning therefore to semiempirical, or nonpredictive models. A novel approach was presented by Granik et al., which is not a priori restricted in concentration domains, presents no adjustable parameters, and is mechanistic, in the sense that it is based on a coupled diffusion model. In this work, we examine the validity of predictive theories of osmosis, by comparison with our new experimental results, and a meta-analysis of literature data.