The largest class of de novo chromosomal rearrangements in Down syndrome are rea(21q21q). Classically, these rearrangements have been termed Robertsonian translocations, implying an attachment of two different chromosome 21 homologues. Additionally, a Robertsonian translocation between two chromosomes 21 cannot be distinguished from an isochromosome composed of genetically identical arms by cytogenetic analyses. Therefore, we have used molecular techniques to differentiate between true Robertsonian translocations and isochromosomes. Samples were obtained from 12 probands, ascertained for de novo rearrangements between homologous chromosomes 21 [11 rea(21q21q) and I rea (21;21)(q22;q22)], their parents (n = 24) and available siblings (n = 7). The parental origins of the de novo rearrangements were assigned using molecular and cytogenetic analyses. Although not statistically significant, there was a two-fold increase in the number of paternally derived de novo rearrangements (n = 8) as compared with maternally derived rearrangements (n = 4). To distinguish between rob(21q21q) and i(21q), we used restriction fragment length polymorphisms (RFLPs) spanning the length of chromosome 21. Using all informative and partially informative RFLPs, we used the method of maximum likelihood to assign the most likely rearrangement definition (i or rob) and parental origin in each family. The maximum likelihood estimates indicated that all rearrangements tested (n = 8) were isochromosomes. C-banding revealed two centromeres in three cases indicating that a U-type exchange occurred between sister chromatids in these rearrangements. Our results suggest that the majority of de novo rea(21q21q) are isochromosomes derived from a single parental chromosome 21. 1987;Priest et al. 1988). In the largest study to date, Grasso et al. (1989) demonstrated that four of six rea(21q21q) were i(21q). We report a study of 12 individuals with chromosome 21 rearrangements and provide evidence for isochromosomes in eight probands using RFLPs for chromosome 21 and maximum likelihood estimation.