Objective To illustrate the impact of intake-related bias in FFQ and 24 h recall (24hR), and correlated errors between these methods, on intake–health associations. Design Dietary intake was assessed by a 180-item semi-quantitative FFQ and two 24hR. Urinary N and urinary K were estimated from two 24 h urine samples. We compared four scenarios to correct associations for errors in an FFQ estimating protein and K intakes. Setting Wageningen, The Netherlands. Subjects Fifty-nine men and fifty-eight women aged 45–65 years. Results For this FFQ, measurement error weakened a true relative risk of 2·0 to 1·4 for protein and 1·5 for K. As compared with calibration to duplicate recovery biomarkers (i.e. the preferred scenario 1), estimating a validity coefficient using this duplicate biomarker resulted in overcorrected associations, caused by intake-related bias in the FFQ (scenario 2). The correction factor based on a triad using biomarkers and 24hR was hampered by this intake-related bias and by correlated errors between FFQ and 24hR, and in this population resulted in a nearly perfect correction for protein but an overcorrection for K (scenario 3). When the 24hR was used for calibration, only a small correction was done, due to correlated errors between the methods and intake-related bias in the 24hR (scenario 4). Conclusions Calibration to a gold standard reference method is the preferred approach to correct intake–health associations for FFQ measurement error. If it is not possible to do so, using the 24hR as reference method only partly removes the errors, but may result in improved intake–health associations.