Surface soil moisture content (SMC) is known to impact soil reflectance at all wavelengths of the solar spectrum. As a consequence, many semi-empirical methods aim at inferring SMC from soil reflectance, but very few rely on physically-based models. This article presents a multilayer radiative transfer model of soil reflectance called MARMIT (multilayer radiative transfer model of soil reflectance) as a function of SMC given on a mass basis and a method called MARMITforSMC to estimate it from soil reflectance spectra. This model depicts a wet soil as a dry soil covered with a thin film of water. It is used to assess SMC over seven independent laboratory datasets gathered from the literature. A learning phase is required to link the thickness of the water film with the SMC. For that purpose, a sigmoid function, the parameters of which are related to soil physical and chemical properties such as porosity, grain size and mineralogy composition, is fitted. SMC can be inferred with good accuracy (RMSE ≈ 3%) if the learning step is applied soil by soil. The link between SMC and water thickness actually depends on soil texture and chemical composition. If the soils are divided into classes and if the learning phase is applied to a class, the RMSE slightly increases up to 5%. Finally, MARMITforSMC provides lower RMSE than any other existing semi-empirical or physically-based method.