Previous studies on the relationships between trace metal accumulation in plants and related soil properties have commonly assumed spatial stationarity by using traditional (global) regression techniques such as ordinary least squares regression (OLS). In this study, geographically weighted regression (GWR) was used to explore the relationships between the Cu concentration in rice (Oryza sativa L.) grain and a set of perceived soil properties, including total Cu in the soil, pH, and soil organic matter in Jiaxing, China. It was found that GWR performed much better than OLS in characterizing the relationships, in terms of the coefficient of determination (R2), corrected Akaike information criterion, ANOVA test, sum of squared residuals, and spatial autocorrelations of residuals. The GWR analysis showed that the relationships between grain Cu and the three related soil properties were not constant across space and there was great spatial non‐stationarity. Results also indicated that GWR could reveal the spatial variations of the relationships ignored by OLS and thus could be used to explain the local causes of Cu accumulation in rice grain. This study demonstrated the advantages of local spatial modeling techniques in modeling the accumulation of hazardous materials in soil–plant systems at regional scales.