Case series of infants with certain birth defect patterns and putative teratogenic exposures should be interpreted with caution since the presence of birth defects and the exposure among the same infants could be entirely due to chance. In the absence of other epidemiologic data, the plausibility for a causal association is strengthened by 1) rarity of the defect pattern, 2) rarity of the exposure in the population, 3) small source population, 4) short time period for the study, and 5) biologic plausibility for the association. These concepts are illustrated using case reports of putative teratogenicity of cocaine and etretinate. In the presence of epidemiologic data, the concept of attributable fraction in exposed (AFE) can be used to evaluate the likelihood that the defect pattern among infants with a particular exposure is attributable to the exposure. This quantity is related to the strength of the epidemiologic association between the defect pattern and the exposure, as measured in terms of relative risk R (or odds ratio), and is equal to (R-1)/R. Even for strong teratogens such as maternal diabetes and isotretinoin, where R is about 7, in more than 14% (1-AFE) of exposed infants with birth defects, the pattern of defects is not attributable to the exposure. Furthermore, AFE can be used to "correct" crude measures of sensitivity (the proportion of exposed malformed infants with a defect pattern attributable to the exposure) and positive predictive value (the proportion of malformed infants who have the exposure and have the defect pattern attributable to the exposure).(ABSTRACT TRUNCATED AT 250 WORDS)