Flux-level-fixup (FF) coarse-mesh finite difference (CMFD) (FF-CMFD), which increases numerical stability during nonlinear iterations for the SP3 advanced nodal method, is proposed as an improved CMFD implementation. In contrast to the scalar flux that appeared in the advanced nodal method with diffusion theory, the second flux moment ϕ2 in the SP3 method could take a very small value since it can take both positive and negative values in a node. This is a root cause of inefficiency of the SP3 advanced nodal method when conventional CMFD acceleration is directly applied. In the proposed FF-CMFD method, a constant value is added to the second flux moment ϕ2 to fix up its value to a sufficiently large positive value for stable numerical calculations. The efficiency of the FF-CMFD method is verified through benchmark calculations.