An advantage of multicenter cohort studies on diet and cancer is that these may include populations over a wide range of dietary exposure. With some simplifying assumptions, the information from such multicenter studies may be divided into 1) estimated relationships within each of the separate cohorts, between individual-level measurements of dietary exposure and disease outcome, and 2) an estimated between-cohort relationship, between the mean intake measurements and mean incidence rates. Errors in the dietary exposure measurements may lead to different amounts of bias in each of these estimated relationships, in particular when dietary questionnaire methods cannot easily be standardized. A calibration approach can be used to adjust for such differences in bias. This will improve the relative weighting of within- and between-cohort components of evidence for a diet-disease association.