Case-control differences in prediagnostic serum levels of retinol, beta-carotene, vitamin E, and selenium are compared for 10 cancer sites in 10 study populations. For all four nutrients, the majority of results showed lower levels among persons who subsequently became cases than among controls. Low levels of beta-carotene were most likely to be associated with subsequent cancer, but there were marked differences by cancer site. The results indicate that it is unlikely that any of these serum micronutrients are associated with protection against carcinogenesis at all sites. A plea is made for greater emphasis on replication of results, for reporting findings for all sites no matter how small the number of cases may be, and for keeping constantly in mind the fact that observational associations are not necessarily causal in nature.