The ability to recognize kin based on genetic markers has been widely proposed as a mechanism to facilitate altruistic behaviour and inbreeding avoidance. Siblings are an important group of relatives to discriminate from unrelated individuals but present a problem, because siblings can share 0, 1 or 2 alleles at any single recognition locus. Here, we present a Bayesian model of kin recognition that defines the potential for genotypic information to convey kinship. Under the direct comparison model, where the signaller’s genotype is compared with that of the receiver, the odds ratio that a pair of individuals were siblings was substantially increased if they shared both alleles at a single locus, but only a minority of siblings were recognized; increasing the number of recognition loci used could not increase both the odds ratio and the proportion of siblings recognized. A maternal comparison model, where the signaller’s genotype is compared with that of the receiver’s mother, performed poorly when only a single recognition locus was considered, but became increasingly effective with more recognition loci. Nevertheless, incorporating partial-matching information across multiple, independent loci are likely to be difficult. Further empirical work needs to establish the mechanistic basis of genetic kin recognition used by different taxa.