Characterizing the interaction between turbulence and radiative processes is necessary for understanding the nocturnal atmospheric boundary layer. The subtle nature of the interaction is exemplified in a phenomenon called the ‘Ramdas paradox’ or the ‘lifted temperature minimum’ (LTM), involving preferential cooling near the Earth's surface. The prevailing explanation for the LTM (the VSN model, Vasudeva Murthy et al (1993 Phil. Trans. R. Soc. A 344 183–206)) invokes radiative exchange in a homogeneous nocturnal atmosphere to predict a large cooling of the near-surface air layers. It is shown here that the cooling predicted by the VSN model is spurious, and that any preferential cooling can occur only in a heterogeneous atmosphere. The underlying error is fundamental, and occurs to varying degrees in a wide class of radiative models, in a flux-emissivity formulation, the VSN model being a prominent example. We, for the first time, propose the correct flux-emissivity formulation that eliminates spurious cooling. Results from field observations and laboratory experiments presented here, however, show that the near-surface radiative cooling is real; near-surface cooling rates can be orders of magnitude higher than values elsewhere in the boundary layer. The results presented include the dependence of the LTM on turbulence, the surface emissivity and the thermal inertia of the ground. It is proposed that aerosols provide the heterogeneity needed for the preferential cooling mechanism. Turbulence, by determining the aerosol concentration distribution over the relevant length scales, plays a key role in the phenomenon. Experimental evidence is presented to support this hypothesis.