Epidemiological studies have consistently reported a reduction in risk for several cancer sites in relation to high consumption of vegetables and fruit. These findings stimulated further research aimed at identifying which compounds in fruit and vegetables are responsible for the reduction in cancer risk. Epidemiological and laboratory studies suggested that some micronutrients, particularly vitamin C, vitamin E, beta-carotene, selenium, magnesium and zinc, could reduce the risk of cancers of the oral cavity, lung, oesophagus and stomach, while dietary fibre was more specifically related to a reduced risk of colorectal cancer. However, the results of large randomized trials on various combinations of vitamins and minerals at supranutritional doses have so far failed to confirm this hypothesis or have found very weak effects. These results should stimulate profound re-thinking of the methods that led to the selection of specific molecules with potential chemopreventive action. From a methodological point of view, little attention has so far been given to four main limitations of nutritional epidemiology, which have direct bearing on the extrapolation of results from foods to food components. Measurements of micronutrient intakes through simple dietary questionnaires and current food composition tables lack precision and specificity. Micronutrient intakes are often highly correlated in nature because micronutrients tend to share the same food sources. Attribution of cancer risk to a single food constituent can be misleading if multicollinearity of dietary variables is not recognized. The etiological meaning of nutritional biomarkers is not straightforward because circulating levels of nutrients reflect not only dietary intake but also complex metabolic regulations. Studies have not considered the physical characteristics of foods, which are important determinants of physiological responses. Understanding the multidimensional nature of diet and of its relationship with different cancers is a major scientific challenge. Epidemiological studies combining detailed dietary questionnaires, appropriate food composition tables, multiple biomarkers, and appropriate statistical methods may provide better measurements of the relationships between cancer risk and specific dietary patterns and therefore contribute to the identification of food components with hitherto unforeseen potential interest for cancer prevention.