Laboratory experiments have been conducted to simulate entrainment in stratiform clouds. In particular, the case of entrainment across a capping temperature inversion and induced by cloud top cooling has been simulated. This geometry is termed interfacial convection, and its physics differ from the more thoroughly studied case of penetrative convection. The dimensionless entrainment rate associated with interfacial convection has been found to vary inversely with a bulk Richardson number over a broad range of Richardson numbers. A dependence of the entrainment rate on the diffusivity of the stratifying agent has also been found. This dependence is explained in terms of Taylor layers. A physical model for the dynamics of interfacial convection is proposed. In the laboratory case, a stably stratified interface separates two fluid layers. Convection is driven in the upper layer by the deposition of radiation near the interface. After sufficient energy has been deposited, buoyant fluid rises and induces formation of entraining cusps at the interface. The spacing between cusps is determined by equating a buoyancy instability timescale with a heating timescale. When the depth of the convecting layer is large compared to the distance separating the cusps, a larger-scale circulation also develops. In such cases, eddies of size comparable to the depth of the convecting layer advect the cusps horizontally. Despite Reynolds numbers that differ by 4 orders of magnitude or more between the laboratory simulation and the real atmosphere, it is argued that the entrainment dynamics are analogous. Dimensionless entrainment rates measured in the laboratory are within 1 order of magnitude of those measured in the atmosphere for a given Richardson number. Thinner Taylor layers and the lack of evaporative effects in the laboratory may account for the difference. ¿ 1998 American Geophysical Union