We present a novel method to detect runs of homozygosity (ROH) from low-coverage genotype data typical for ancient human DNA. ROH are the genetic signature of matings between related parents, and as such, the frequency and length distribution of these blocks can give insight into recent population history and mating patterns. Existing methods identify ROH by scanning for regions that lack heterozygote genotypes, but this strategy frequently fails for ancient individuals: The vast majority of ancient DNA data has low read depth (ă3ˆ), which makes reliable diploid genotype calling infeasible. To overcome this limitation, we make use of linkage disequilibrium information from a panel of modern reference haplotypes using a Hidden Markov Model. Our method scans for long stretches where the read data are consistent with only a single haplotype. When tested on simulated and down-sampled pseudo-haploid data from a targeted set of 1.24 million single nucleotide polymorphisms ("1240k SNPs") widely used in ancient DNA, our implementation robustly works for coverage down to 0.5ˆand can tolerate error rates up to 3%, with high power and low false positive rate for blocks longer than 4 centiMorgans. Therefore, the method can screen a substantial fraction of human genome-wide ancient DNA data for parental relatedness, which will yield new evidence for questions regarding past demography and social organization.