<p>Geographically weighted mean (GWM) is a fundamental geographically weighted descriptive statistic for characterizing spatial heterogeneity of a geographic variable, yet its application has long been constrained by the absence of a principled bandwidth selection approach. In practice, bandwidth choice is often guided by rules of thumb or cross-validation strategies that lack clear interpretation and theoretical grounding. To address this limitation, we here proposed an analysis of variance (ANOVA) based approach of bandwidth optimization for GWM. By conceptualizing bandwidth-defined local neighborhoods as spatial groupings and explicitly quantifying between- and within- neighborhood variability through the F-statistic, the proposed approach identifies the bandwidth at which spatial heterogeneity is most effectively expressed. It demonstrated that bandwidth should not be viewed merely as a smoothing parameter, but rather as a scale-controlling mechanism that determines which spatial structures are emphasized or suppressed in GWM. Evaluations using both simulated datasets with diverse spatial patterns and an empirical case study, the proposed method yields stable and interpretable bandwidth choices, while avoiding the pitfalls of overfitting associated with overly small bandwidths and excessive smoothing induced by overly large ones. Overall, the proposed approach enhances the interpretability and methodological transparency of GWM and offers a generalizable choice to its scale selection, with potential applicability to a wide range of spatial data exploration and modeling tasks.</p>