This question is a fundamental and difficult issue causing confusion in clinical care and research. To delineate complexities arising when conventional approaches may be counterproductive, we examine the following questions in the context of our recently completed Vitamin A Trial showing a reduction in CLD: 1) What should “normal” blood levels indicate? a) Typical values (e.g., mean±2SD for serum retinol) in a referent population (term infants) or in the population of interest (ELBW infants)? or b) Desirable values for this population?--Our choice for the trial. The latter may exceed published “norms” for term infants if vitamin A consumption increases during lung repair in ELBW infants. 2)Are blood levels likely to reflect tissue levels in the population of interest?--Not in our trial, partly because the steroids used to treat sicker infants increase serum retinol levels by reducing tissue stores. 3)Should dosage be adjusted by blood levels in clinical trials and supplements discontinued when levels exceed “normal” values in healthy populations?--Not for our trial. For the above reasons, use of target retinol levels might produce misleading results as well as increase trial complexity and cost. Because desirable levels for ELBW infants are unclear, we provided a specific dose and assessed toxicity and need to discontinue vitamin A for individual infants by clinical exams rather than biochemical assays. 4) Can normal (in the sense of desirable) blood levels be determined in observational studies?--Usually not because confounders may be present. In our trial, serum retinol levels [median(range)] were 29.6(10-105) μg/dl in supplemented infants and 17.8(4-132) μg/dl in controls. In each group, retinol values were twice as high in steroid-treated infants. When our retinol data for the control group were analyzed as in an observational study, values above the median were associated with a worse primary outcome (p=0.05), reflecting the effect of steroids or other confounders. Yet when the two randomized groups were compared, higher retinol values (in the vitamin A group) were associated with a better clinical outcome and no evidence of toxicity (see separate abstract). When confounders are present, a randomized trial is required to determine blood levels or intakes that result in the best outcome. Our trial indicates the need to carefully consider the definition and use of “normal” values in clinical care and research.