Several intellectual "autopsies" have recently reviewed errors in clinical epidemiologic studies of causation, such as the original claim that amyl nitrite "poppers" caused AIDS. The current autopsy was done to determine why it took more than a decade--1942 to 1954--to end an iatrogenic epidemic in which high-dose oxygen therapy led to retrolental fibroplasia (RLF) in premature infants, blinding about 10,000 of them. The autopsy revealed a museum of diverse intellectual pathology. When first noted, RLF was regarded as neither a new disease nor a postnatal effect. In early investigations, the ophthalmologists did not establish explicit criteria for diagnosis and confused RLF with malformations previously seen in full-term infants. Because the patients were not referred until months after birth, the ophthalmologists assumed that the lesion, which resembled an embryologic structure, must have occurred prenatally. Other events suggesting a prenatal cause for RLF were its strong statistical associations with fetal anomalies, multiple gestations, and maternal infections. Although these events were also associated with prematurity, it was ignored when the RLF cases were compared with controls who were mainly full-term infants. The postnatal timing of RLF was eventually recognized when investigators did cohort studies in premature infants and found that RLF could develop in eyes that were normal at birth. As the search for a cause turned to events occurring after birth, statistical associations were produced for agents such as light, vitamins, iron, vitamin E deficiency, and hypoadrenalism. Each study had its own methodologic flaws: controls were missing for light; co-maneuvers were ignored for vitamins and iron; objective diagnosis was not used for vitamin E deficiency; and the research on hypoadrenalism contained biases in susceptibility and detection as well as problems of a competing outcome event. When the role of oxygen administration was first considered, the statistical association with RLF was stronger for vitamin- and iron-therapy than for oxygen. In addition, many investigators were dissuaded by contradictory evidence from institutions in which RLF was either absent despite high-dose oxygen or persistent despite reduced dosage. The contradictory evidence was later regarded as erroneous because of unsatisfactory delivery systems for the oxygen or failure to check the actual oxygen concentrations. An alternative explanatory hypothesis, rejecting the role of high-dose and long-duration oxygen, was the idea that RLF was due to "relative hypoxia", produced by overly rapid weaning from oxygen therapy rather than the duration of oxygen treatment itself.(ABSTRACT TRUNCATED AT 400 WORDS)