Geometric Volume-of-Fluid methods are a popular choice for the simulation of separated two-phase flows (cf. [1] for a recent review). These methods require a geometric piecewise linear reconstruction of the fluid interface in every cell containing both fluid phases. With the Reconstructed Distance Function (RDF) technique [2,3], the interface normals are estimated based on the gradient of a reconstructed distance function. A major challenge during interface reconstruction is that small errors in the normal estimation can result in large errors in the interface curvature, inducing so-called parasitic currents in the simulation. Therefore, improving the RDF calculation is crucial to increase simulation accuracy. We introduce a filter based on the size of the PLIC polygon into the approximation of the RDF, as cells with small PLIC polygons are prone to high errors in interface normal estimation. The method is implemented in OpenFOAM and its effect is examined using canonical test cases for segregated two-phase flows. The results are discussed with regard to curvature accuracy and magnitude of parasitic currents. REFERENCES [1] Marić, T., Kothe, D.B., and Bothe, D. (2020). Unstructured un-split geometrical Volume-of-Fluid methods – A review. Journal of Computational Physics 420, 109695. [2] Cummins, S.J., Francois, M.M., and Kothe, D.B. (2005). Estimating curvature from volume fractions. Computers & Structures 83, 425–434. [3] Scheufler, H., and Roenby, J. (2019). Accurate and efficient surface reconstruction from volume fraction data on general meshes. Journal of Computational Physics 383, 1–23.