Recent discoveries reveal many details about mechanisms of heat transfer in nanoparti-cle–fluid suspensions or nanofluids. Some of the existing mathematical models even allow molecular simulations to be carried out with various phenomena taken into account to fit theoretical data with experimental better. However, so-called one phase or homog-enized models describing the macroscopic behavior of nanofluids as homogeneous fluids with effective parameters sometimes demonstrate a substantial gap with experimental data. Reducing or possibly even removing that gap is one of the directions in which the current theory of nanofluids must be developed. This paper attempts to reduce the gap by proposing a multiscale description of nanofluids. More specifically, we provide separate sets of partial differential equations (PDEs) for carrying out microscale (molecular level, expressed in nm), mesoscale (particle level, expressed in µm) and macroscale (ho-mogenized level, expressed in mm) simulations of heat transfer in nanofluids. Variational convergence between these sets of PDEs at different scales is established by means of the theory of distributions and weak convergence of distributions. The mesoscale model, corresponding to a fluid medium with point inhomogeneities at the locations of particles, allows a quadratic dependence of the thermal conductivity with respect to the particles’ 1 volume fraction to be derived. As the numerical evaluation and comparison with available experimental data shows, the proposed model is in much greater agreement with experiment than the existing fundamental linear models.