Studies that explore associations between the local food environment and diet routinely use global regression models, which assume that relationships are invariant across space, yet such stationarity assumptions have been little tested. We used global and geographically weighted regression models to explore associations between the residential food environment and fruit and vegetable intake. Analyses were performed in 4 boroughs of London, United Kingdom, using data collected between April 2012 and July 2012 from 969 adults in the Olympic Regeneration in East London Study. Exposures were assessed both as absolute densities of healthy and unhealthy outlets, taken separately, and as a relative measure (proportion of total outlets classified as healthy). Overall, local models performed better than global models (lower Akaike information criterion). Locally estimated coefficients varied across space, regardless of the type of exposure measure, although changes of sign were observed only when absolute measures were used. Despite findings from global models showing significant associations between the relative measure and fruit and vegetable intake (β = 0.022; P < 0.01) only, geographically weighted regression models using absolute measures outperformed models using relative measures. This study suggests that greater attention should be given to nonstationary relationships between the food environment and diet. It further challenges the idea that a single measure of exposure, whether relative or absolute, can reflect the many ways the food environment may shape health behaviors. absolute exposure; fruit and vegetable intake; geographically weighted regression; local food environment; nonstationarity; relative exposure Abbreviations: AICc, corrected Akaike information criterion; GWR, geographically weighted regression; ORiEL, Olympic Regeneration in East London.Over the last decade, an extensive body of research has investigated how the local food environment may be related to dietary behaviors (1). A large majority of studies have used global regression to model the association between exposure to either healthy or unhealthy food environments and diet-related outcomes (2, 3). However, reviews have highlighted the lack of consistency in findings (2, 3), with associations for the same type of relationship being positive (4-10), negative (11), or nonexistent (12-18).By using global regression models, researchers have implicitly relied on the assumption of a stationary relationship; that is, parameter estimates describe what is assumed to be an invariant relationship across space. However, public health researchers have begun to challenge the stationarity assumption. Using spatial regression modeling, such as geographically weighted regression (GWR), a technique that allows for spatial variations in parameter estimates (19,20), investigators have highlighted variations in associations across space between a range of environmental exposures and outcomes such as diet (21), obesity (22-27), active transportation (28), and bi...