Within the framework of the Generation IV Sodium-cooled Fast Reactor (SFR) R&D program of CEA (French Commissariat à l’Energie Atomique et aux Energies Alternatives), reactor behavior in case of severe accidents is assessed. Such accidents are usually simulated with mechanistic calculation tools (such as SAS and SIMMER). As a complement to these codes, which give reference accidental transient, a new physico-statistical approach is currently followed by CEA; its final objective being to derive the variability of the main results of interest for the safety. This approach involves a fast-running description of extended accident sequences coupling physical models for the main phenomena with advanced statistical analysis techniques. It enables to perform a large number of simulations in a reasonable computational time and to describe all the possible bifurcations of the accident transient. In this context, this paper presents a physical tool (models and results assessment) dedicated to the beginning of the primary phase of an Unprotected Loss Of Flow accident (i.e. before the first pin degradation). This study focuses on the sodium thermalhydraulic behavior during this phase, because according to some previous boiling tests, a stabilized boiling flow could be achieved at the top of some subassemblies, thus preventing pins complete dry-out and consequent core degradation. Experiments that demonstrated this possible boiling stabilization were carried out at constant power. However, in case of an unprotected accident, power will vary according to reactivity feed-back effects. This physical tool is described before presenting the comparison of its results with experimental tests results (static and dynamic aspects) for the thermalhydraulic behavior and with mechanistic SIMMER code results for the global core evolution (including neutronics) during an ULOF transient. Moreover, the flow evolutions obtained with this tool are demonstrated to be in good agreement with the Ledinegg’s quasistatic theory regarding boiling stabilization possibilities or flow excursion. This tool is demonstrated to be capable of reproducing the magnitude of mass-flow rate, reactivity, power evolution during an ULOF with a discrepancy of less than 3% regarding the same transient simulated with SIMMER. It will be used for safety-informed design and stability analyses of fast reactor systems, allowing to emphasize main dominant phenomena and significant trends for safety assessment. In the future, this physical tool, associated to statistical treatments of the effect of uncertainties, will enable large sensitivity analysis studies.