Despite widespread acceptance of the concept of very low birth weight (VLBW), i.e., birth weight of less than or equal to 1,500 g, VLBW infants represent an extremely heterogeneous group of newborns, including those with very immature gestational age and those who are more mature but extremely growth retarded. To demonstrate how use of the VLBW rubric can lead to confounding bias that is not only large in magnitude but impossible to control satisfactorily, the authors divided 640 consecutive live neonates born in the Royal Victoria Hospital, Montreal, Canada, from 1978 to 1987 into two overlapping groups: a VLBW cohort (birth weight, 500-1500 g; n = 573) and a gestational age cohort (gestational age, 23-30 completed weeks; n = 466). Variation in growth status by gestational age was much more uniform in the 23- to 30-week cohort. Thus, although mean birth weight was similar in the 500- to 1,500-g and 23- to 30-week cohorts (1,055 vs. 1,064 g), the 500- to 1,500-g cohort was more mature (mean gestational age, 28.8 vs. 27.8 weeks; upper range, 39.7 vs. 30.9 weeks) and had twice the rate of intrauterine growth retardation (25.7 vs. 11.5%). These differences in maturity and growth resulted in a misleading protective effect of intrauterine growth retardation against in-hospital death in the 500- to 1,500-g cohort (crude odds ratio = 0.55 (95% confidence interval 0.36-0.83] and a greater discrepancy in maturity between cesarean- and vaginally delivered infants (3.1 vs. 1.5 weeks) in the 500- to 1,500-g vs. 23- to 30-week cohorts. These differences arise from inextricable confounding of growth status and maturity in the 500- to 1,500-g cohort, the most mature infants also being the most growth retarded. The removal of well-grown infants with birth weights of greater than 1,500 g from the VLBW cohort leads to a progressively distorted spectrum of growth with advancing gestational age and an artifactual blunting of the beneficial effects of increasing maturity. The authors suggest that whenever fetal growth is an important exposure, outcome, or confounding variable, epidemiologic studies of extremely small or immature newborns should be based on gestational age rather than the VLBW criterion.