The flamenco (flam) locus, located at 20A1-3 in the centromeric heterochromatin of the Drosophila melanogaster X chromosome, is a major regulator of the gypsy/mdg4 endogenous retrovirus. In restrictive strains, functional flam alleles maintain gypsy proviruses in a repressed state. By contrast, in permissive strains, proviral amplification results from infection of the female germ line and subsequent insertions into the chromosomes of the progeny. A restrictive/permissive polymorphism prevails in natural and laboratory populations. This polymorphism was assumed to be maintained by the interplay of opposite selective forces; on one hand, the increase of genetic load caused by proviral insertions would favor restrictive flam alleles because they make flies resistant to these gypsy replicative transpositions and, on the other, a hypothetical resistance cost would select against such alleles in the absence of the retrovirus. However, the population cage data presented in this paper do not fit with this simple resistance cost hypothesis because restrictive alleles were not eliminated in the absence of functional gypsy proviruses; on the contrary, using 2 independent flam allelic pairs, the restrictive frequency rose to about 90% in every experimental population, whatever the pair of alleles and the allelic proportions in the initial inoculum. These data suggest that the flam polymorphism is maintained by some strong balancing selection, which would act either on flam itself, independently of the deleterious effect of gypsy, or on a hypothetical flanking gene, in linkage disequilibrium with flam. Alternatively, restrictive flam alleles might also be resistant to some other retroelements that would be still present in the cage populations, causing a positive selection for these alleles. Whatever selective forces that maintain high levels of restrictive alleles independently of gypsy, this unknown mechanism can set up an interesting kind of antiviral innate immunity, at the population level.