Over past decades, analyses of transient processes and accidents in a nuclear power plan t have been performed, to a significant extent and with an admirable success, by means of so called system codes, e.g. RELAP5, CATHARE, ATHLET codes. These computer codes, based on a multi-fluid model of two-phase flow, provide an effective, one-dimensional description of the coolant thermal-hydraulics in the reactor system. For some components in the system, wherever needed, the effect of multi-dimensional flow is accounted for through approximate models. The later are derived from scaled experiments conducted for selected accident scenarios. Increasingly, however, we have to deal with newer and ever more complex accident scenarios. In some such cases the system codes fail to serve as simulation vehicle, largely due to its deficient treatment of multi-dimensional flow (in e.g. downcomer, lower plenum). Enter Computational Fluid Dynamics (CFD). Based on solving Navier-Stokes equations, CFD codes have been developed and used, broadly, to perform analysis of multi-dimensional flow, dominantly in non-nuclear industry and for single-phase flow applications. Although not always straightforward, CFD codes can be, and have been, used to analyze thermo-fluid processes in a certain component of the reactor system at a well-defined point during the accident progression. It is natural to think that CFD codes provide the much-needed complementary capability to the system codes. Furthermore, due to the CFD excessive demand on computational resources, ideas were proposed, and attempts were reported in the literature, to use a coupled system-to-CFD code to maximize the benefit of both tools. Easy as it might sound, progress in this area has been sluggish. In this paper, we take a close look at the progress in coupled system-to-CFD code analyses, including coupling algorithms, their implementation and performance. Tackling thermo-fluid dynamics at largely different scales, system codes and CFD codes employ different models and governing equations. This notion led us to the idea to examine the system-to-CFD coupling in the language of multiscale simulations. As a theoretical framework, we bring to bear the heterogeneous multiscale method pioneered by E and Engquist and problem classification offered by E et al.. Viewing system-to-CFD coupling as a multiscale problem, the ultimate objective of the present effort is to define requirements for models and numerical methods, and develop suggestions on a coupling strategy that ensures robust and effective generation and transfer of information between scale-specific simulations (system and CFD).