To the Editor In their data analysis from 2 observational cohort studies, Feskanich et al1 found that greater milk consumption during adolescence was not significantly associated with a lower risk of hip fracture in older women, but was weakly and significantly associated with a greater risk of hip fracture in older men. They also reported that the weak positive association in men was partially mediated by greater height. Their data analysis is limited by not including many important confounding variables that are associated with increased fracture risk. These variables include the risk of falls (which itself is multifactorial, but can be quantified), family history of fracture or osteoporosis, concurrent medical disorders associated with osteoporosis or fractures, carbonated beverage consumption (past and present), and the use of certain drugs (eg, proton pump inhibitors, anticonvulsants, thyroid hormone). Patients with a history of hip fracture were excluded from the analysis, but history of any fracture (data not reported by the authors) is associated with an increased risk of future fractures; this should be considered a confounding variable. The investigators acknowledge the possibility of error in retrospective reporting of milk consumption, but their discussion of this limitation minimizes the significance of this problem. Correlations between documented consumption and recalled intake are weak. This is one of several problems that are a serious pitfall for nutritional research.2 Feskanich et al characterize greater height as an independent risk factor for hip fracture and they consider this a possible intermediate variable in their analysis. A critical appraisal of published cohort and case-control studies pertaining to height and fracture risk (2 studies referenced by the authors and multiple studies referenced in articles by Hemenway et al3 and Compston et al4) reveals mixed findings from these studies. Not all studies confirmed an association and positive effects were uniformly small. The relationship between height and fracture location (hip, spine, or limb) also differed in some analyses. The effect of sex noted by Feskanich et al conflicts with findings from some of these studies. Virtually all of these studies (including the study by Feskanich et al) suffer from confounding variables and bias. Weak and inconsistent associations, taken from observational epidemiologic studies, are likely to be false and attributable to various sources of bias.2,5 Because of method limitations in their study design and data analysis, the conclusions of Feskanich et al cannot be considered valid.