Usage of 3D roughness parameters measurements requires knowledge of uncertainty causes in order to design proper measurement procedures. Even on apparently homogeneous machined surfaces, large local variations in 3D roughness parameters have previously been reported. This paper introduces and tests a concept for optimized sampling based on local-, and global topography analysis, using measurements of the Sa parameter on a large machined component as a practical example. It is shown that more than 40 measurements of Sa may be needed to attain a stable value, but that choice of an appropriate sampling strategy may reduce this requirement considerably. The results point to a possible route to minimize required measurements and contribute to the development of the "best-practice" when using 3D surface structure metrology.