undiagnosed disease often results in changes of diet The assumption that diet does not change markedly over time is almost certainly errone- ous. One example from the Nurses Health Study showed that 77 of women changed their dietary fiber intake by two or more quintiles ver only 6 years(Fuchs et al. 1999). Interpretation of those results can provide only two reasonable conclusions that are not mutually ex clusive: that the diet changes considerably over time or that the FFQis simply not reproducible enough to provide certainty about intakes. So that leaves readers with only prospective cohort studies that use routine evaluations of diet every few years as the studies to consider even though this distinction is rarely applied in meta-analyses of this area. is important also to realize the imitations of assessing dietary intake. This function is dependent upon the accuracy ofboth food composition databases and the dietary survey instnument.Given the huge number o food products in the marketplace. it is impossible to keep up with changes For example the USDA food composition database, considered the largest and most accurate in the world, has 8600 foods but the American food supply has up to 100.000 distinct food products many of which are in the marketplace for a short time or have their formulas changed periodically. Some products in the USDA database have no been updated for many years and some compositions are imputed from typical formulations. The USDA database is also used in many other countries that do not have analyses oftheirown food supplies,po- tentially making the accuracy of nutrient information from studies in her countries highly suspect. To enable analysis of large numbers of dietary surveys by machine- scoring food-frequency questionnaires(FFQs) have become the dard means of evaluating food consumption, although a few studies have used multiple 24-h recalls or multi-day food diaries. Some FFQs do not ask about portion size but only frequency of consumption. Be- cause people with different weights and levels of physical activity have a very large difference in energy requirements, food and nutrient intake in most FFQs are adjusted for total energy intake. While this is logical, a number of studies have dearly shown that the FFQis poor a determining energy intake. If there is little confidence in the divisor(energy), it is impossible to attribute much certainty to the numerator(a specific food or nutrient) In addition, the standard validation" of n FFQ has been to determine how well it correlates with intake as mea sured by 24-h recall. Since recalls are usually t25roftrue intake as de- termined by doubly labeled water, and FFQs correlate with recalls on