A cladistic analysis was conducted on restriction site variation in the inverted repeat region of the chloroplast DNA of 34 taxa, including 22 genera of Hamamelididae, nine genera of Rosidae, and one genus of Magnoliidae (sensu Cronquist). Parsimony analysis of 45 informative characters resulted in four equally parsimonious cladograms. In all trees, Hamamelididae were polyphyletic, with the rosid taxa nested within lower hamamelids and higher hamamelids in turn nested within this rosid group. The higher Hamamelididae (excluding Leitneria but including Betulaceae, Casuarinaceae, Fagaceae, Juglandaceae, Myricaceae, Nothofagaceae) were monophyletic in all of the most-parsimonious trees, and were always a sister group of Malus (Rosaceae). Within the higher hamamelid group, Nothofagus was found to be the sister taxon of the remainder of the taxa, supporting recognition of the family Nothofagaceae. The genera of Fagaceae s. str. formed a monophyletic group and were a sister group of the remainder of the higher hamamelids excluding Nothofagus. The sister group of the rosid-higher hamamelid clade in all trees was a clade that included Liquidambar, Cercidiphyllum, and representative genera of Hamamelidaceae s. str. The analysis showed Euptelea to be the basal taxon of lower hamamelids, whereas Trochodendron was identified as sister taxon to a larger clade comprised of Hamamelidaceae, Cercidiphyllum, rosids, and higher hamamelids. Cercidiphyllum was shown to be closely related to Hamamelidaceae s.l., whereas Platanus was placed several nodes away from this clade. This analysis supports recent contentions that Hamamelididae sensu Cronquist is not monophyletic. Our results contradict recent suggestions based on fossil leaves that Fagaceae are derived directly from a platanoid ancestor. The dicotyledonous subclass Hamamelididae has been the focus of considerable attention by systematists in recent years (see Crane and Blackmore 1989 for a bibliography). As circumscribed by Cronquist (1981, 1988; our Table 1), and in essentially similar form by Takhtajan (1980, 1987) the Hamamelididae are highly polytypic, comprising a diverse assemblage of families that lack unique characters and share many putatively diagnostic features with other groups, such as Rosidae (sensu Cronquist 1981). Such groups are suspect, and often are found to be polyphyletic or paraphyletic when carefully analyzed using cladistic methods. The hypothesis that various Hamamelididae may be more closely related to some rosid families is supported by numerous recent morphological, anatomical and palynological studies of Hamamelididae and Rosidae (Crepet and Nixon 1987; Dickison 1989; Ehrendorfer 1977, 1989; Hickey and Doyle 1977; Hickey and Taylor 1991; Hickey and Wolfe 1975; Hufford 1992; Nixon 1985,1989; Walker and Doyle 1975; Wolfe 1989; J. Wolfe et al. 1975). The few studies that have proposed explicit hypotheses of relationship based on modern cladistic concepts and methods suggest that the broadly defined Hamamelididae is polyphyletic, with some elements more closely related to rosids than to other hamamelids (Hufford 1992; Nixon 1984, 1989; Nixon and Manos, unpubl. data). Of particular interest is one group of families often called the higher Hamamelididae (e.g., Crane and Blackmore 1989; Table 1). The members of this group mostly (but not consistently) share wind-pollination and granular-columellate pollen walls and reduced apertures. Based on morphological cladistic analyses (Hufford 1992; Nixon 1984, 1989), higher hamamelid families are derived from families commonly considered to be basal within Rosidae, such as Cunoniaceae or Brunelliaceae. In turn, the lower rosid families are nested within the lower hamamelid families, as a sister group of a lineage including modern Hamamelidaceae. To address these hypotheses and provide