In nutritional epidemiology, our goal is to understand how what we usually eat and drink influences our health and disease risk. However, usual intake is typically not observable in free-living humans. Therefore, nutrition researchers often rely on self-reported dietary intake data elicited using dietary recalls, food records, and food frequency questionnaires, which are subject to measurement error that can lead to misleading findings. Systematic error comes about due to recall and reactivity biases, misalignment between cognitive demands and capacities, and social desirability biases. Systematic error impacts food frequency data to a greater extent than recall or record data, but the latter are affected to a larger degree by random error, driven by day-to-day variation in intake that impacts our capacity to estimate usual intake. Advances in measurement error modelling have helped to characterize the error in dietary intake data collected using different tools and informed strategies to mitigate this error, lending confidence to the findings of diet and disease studies. Ongoing challenges include extending measurement error models to address the complexity of eating patterns.