The wood anatomy of the Rosaceae is surveyed and analysed, based on the study of 280 species (c. 500 specimens) belonging to 62 genera from different parts of the world. Eighteen wood anatomical characters have been used for a phenetic and phylogenetic classification. In the phenetic classification, 12 groups are recognised and compared with Hutchinson’s tribes. Groups I-V accommodate a mixture of representatives from Spiraeoideae and Rosoideae genera (or tribes); Groups VI-VII comprise the Maloideae plus Cercocarpus; Groups VIII-IX are composed mainly of the tribe Quillajeae, but include Prinsepia, and exclude Exochorda; Groups X-XII represent the Prunoideae. All woody tribes recognised by Hutchinson encompass few closely similar wood anatomical groups except the Cercocarpeae. The Quillajeae, characterised by presumably primitive macromorphological characters, appear to be quite derived wood anatomically, and, according to the phylogenetic analysis, should be included in the Prunoideae or be recognised as a separate subfamily close to the Prunoideae. The wood anatomical variation patterns support a broad generic delimitation in the Maloideae, and a narrow one in the Prunoideae. The wood anatomical evidence also suggests that Cercocarpus be moved to the Maloideae, and that Sorbaria is the most derived representative in the Spiraeoideae/ Rosoideae. Prinsepia, although distinct in its wood anatomy from the Prunus alliance, still forms part of a monophyletic group together with the latter. Based on wood anatomical evidence, the subdivision of the family and the origin of the Maloideae are discussed. Wood anatomically the Rosaceae are isolated within the order Rosales. It is hypothesised that the Rosaceae might have originated in East Asia instead of in the tropics. Individual generic descriptions, a family description and a tentative wood anatomical key to the genera studied are also presented.