Kin–recognition systems have been hypothesized to exist in humans, and adaptively to regulate altruism and incest avoidance among close genetic kin. This latter function allows the architecture of the kin recognition system to be mapped by quantitatively matching individual variation in opposition to incest to individual variation in developmental parameters, such as family structure and co–residence patterns. Methodological difficulties that appear when subjects are asked to disclose incestuous inclinations can be circumvented by measuring their opposition to incest in third parties, i.e. morality. This method allows a direct test of Westermarck's original hypothesis that childhood co–residence with an opposite–sex individual predicts the strength of moral sentiments regarding third–party sibling incest. Results support Westermarck's hypothesis and the model of kin recognition that it implies. Co–residence duration objectively predicts genetic relatedness, making it a reliable cue to kinship. Co–residence duration predicts the strength of opposition to incest, even after controlling for relatedness and even when co–residing individuals are genetically unrelated. This undercuts kin–recognition models requiring matching to self (through, for example, major histocompatibility complex or phenotypic markers). Subjects' beliefs about relatedness had no effect after controlling for co–residence, indicating that systems regulating kin–relevant behaviours are non–conscious, and calibrated by co–residence, not belief.