The early stage of silicon heteroepitaxial growth and germanium homoepitaxial growth on Ge(001) has caused a discrepancy between experimental and theoretical work. Previously a dimer configuration was identified experimentally, which theoretically has been predicted to be unfavorable: the D dimer. Upon careful reinvestigation this cluster turns out to be not a two-atom, but a three-atom cluster. The three-atom cluster of Ge or Si on Ge(001) is shown to differ from a C dimer, a small epitaxial island (BD) or the three-atom cluster of Si on Si(001).