The paper discusses the primary and secondary endings of the Indo-European middle. It is suggested that, rather than being a local (Italo-Celtic) innovation, the r-endings of the middle voice represent a Proto-Indo-European archaism. Numerous middle forms containing the element -r- are found not only in the Tocharian languages, but also in most of the Anatolian languages attested in the second millennium BC (including Hittite, Palaic, Cuneiform Luvian and Hieroglyphic Luvian). Other Indo- European languages (including Greek and Indo-Iranian) display a zero marker, whereas the oldest Hittite texts attest the primitive feature -t-. The Old Hittite middle marker *-ti, it is claimed, was more archaic than its late variants *-ri as well as *-i. The original primary middle endings in non-Anatolian Indo-European should be reconstructed as follows: 1 sg. pres. *-mh2eŘi, 2 sg. *-sh2eŘi, 3 sg. *-toŘi, 1 pl. pres. *-mesdhh2oŘi, 2 pl. *-sdh(u)u̯ eŘi, 3 pl. *-ntoŘi for transitive verbs and 1 sg. *-h2e/oŘi, 2 sg. *-th2eŘi, 3 sg. *-oŘi, 1 pl. *-medhh2oŘi, 2 pl. *-dh(u)u̯ eŘi, 3 pl. *-roŘi for intransitive verbs. The Indo-European phoneme *Ř seems to be a refl ex of a Proto-Indo-European (i.e. Indo-Hittite) dental stop *Ď, probably identical with the Indo-European dental spirant *đ.