Numerical simulations of the primary phase of a hypothetical core disruptive accident are currently based on a multiple-channel approach, which requires that subassemblies or groups of subassemblies be represented together as independent channels. Generally, a single-pin treatment is used to model the channel fuel pins. The limitation of this simplified approach should be assessed because it can affect voiding and melting patterns that in turn may influence reactivity insertions and power history. In the same manner, the single-pin hypothesis may introduce important biases in the prediction of can-wall thermal ablation. Radial propagation of the degradation and subsequent accident consequences may thus be affected. To improve the safety assessment of sodium fast reactors, two-dimensional effects are investigated using a multiple-pin model. Numerical results for a severe accident transient show that the current methodology is nonconservative and predicts the onset of sodium boiling with a delay. A two-node radial meshing of the subassembly is preferred for treating the peripheral ring of fuel pins separately from the rest of the pins. This treatment would allow overcoming the previous issue and give more accurate initiating phase simulations.