The stable isotope stratigraphy of river- and lake-ice archives winter hydroclimatic conditions, and can potentially be used to identify changing water sources or to provide important insights into ice formation processes and growth rates. However, accurate interpretations rely on known isotopic fractionation during ice growth. A one-dimensional diffusion model of the liquid boundary layer adjacent to an advancing solid interface, originally developed to simulate solute rejection by growing crystals, has been used without verification to describe non-equilibrium fractionation during congelation ice growth. Results are not in agreement, suggesting the presence of important uncertainties. In this paper we seek validation of the diffusion model for this application using large-scale laboratory experiments with controlled freezing rates and frequent sampling. We obtained consistent, almost constant, isotopic boundary layer thicknesses over a representative range of ice growth rates on both quiescent and well-mixed water. With the 18O boundary layer thickness from the laboratory, the model successfully quantified reduced river-ice growth rates relative to those of a nearby lake. These results were more representative and easier to obtain than those of a conventional thermal ice-growth model. This diffusion model validation and boundary layer thickness determination provide a powerful tool for interpreting the stable isotope stratigraphy of floating ice. The laboratory experiment also replicated successive fractionation events in response to a freeze-thaw-refreeze cycle, providing a mechanism for apparent ice fractionation that exceeds equilibrium. Analysis of the composition of snow ice and frazil ice in river and lake cores indicated surprising similarities between these ice forms. Published in 2002 by John Wiley & Sons, Ltd.