1325 R.DULYMAMODE 1 , P. F. CANNON 2 AND A.PEERALLY 3  Faculty of Science, University of Mauritius, Re  duit, Mauritius  CABI Bioscience, Bakeham Lane, Egham, Surrey TW20 9TY, UK  University of Mauritius, Re  duit, Mauritius Astrocystis fimbriata, A. rarissima and A. cepiformis spp. nov. are described and illustrated from dead leaves of Pandanus. Their affinities with related taxa are discussed, especially with Astrocystis species on Palmae and with Rosellinia as interpreted in recent studies. One species has stromata which develop from a mat of superficial hyphae, which has not previously been reported for Astrocystis. Studies of collections made during the period 1995–6 as part of a preliminary assessment of fungal diversity in the remaining forest reserves of Mauritius are revealing many novel species. Three stroma-producing species are here assigned to Astrocystis Berk. & Broome, a probable close relative of Rosellinia de Not. Rosellinia was erected by De Notaris (1844) with R. aquila as the type species. Useful modern treatments of the genus include those by Dargan & Thind (1979), Francis (1985), Petrini (1993) and Læssøe & Spooner (1994). The last two papers explore the distinctions between Rosellinia and Astrocystis, a genus first described for a single bamboo-inhabiting species with a prominent stellate structure surrounding the stroma. This is composed of blackened outer layers of the host epidermis which are pushed back as the stroma matures. Læssøe & Spooner (1994) considerably increased the number of species assigned to this last genus, and included three species on palms (the name of one of which was not formally combined into Astrocystis). Not all experts on the Xylariaceae agree that Astrocystis is distinct from Rosellinia. In particular, Ju & Rogers (1990) concluded that the two taxa were synonymous, although much of their argument relied on acceptance of R. bambusae Henn. as a Rosellinia rather than an Astrocystis, as advocated by Læssøe & Spooner (1994). Læssøe & Spooner considered that the two genera could be separated using the following characters. The form and size of the apical ring in Rosellinia is a ‘ massive, typically barrel-shaped apical apparatus ’, while in Astrocystis it is small and cylindrical or funnel-shaped. The ascus stalk in Rosellinia tends to be longer than in Astrocystis. Superficial hyphae surrounding stromata (referred to by them as a ‘ false ’ subiculum) are often present in Rosellinia, but were stated to be absent in Astrocystis. Finally, the two genera have different anamorphic forms (Geniculosporium and Dematophora for Rosellinia, Acanthodochium for Astrocystis). Læssøe & Spooner (1994) agreed with Ju & Rogers (1990) that the stellate or lobed structures surrounding the stroma were likely to be more a feature of the host epidermal morphology than of the fungus, and they therefore included a number of species in Astrocystis without this character. Other bamboo-inhabiting genera (e.g. Astrosphaeriella Syd. ; Hawksworth, 1981) exhibit such stellate structures around the erumpent ascomata. The assignment of the three species described here to a genus is problematic. They all have asci with relatively small flared apical rings and short stalks, which are typical of Astrocystis in the sense of Læssøe & Spooner. One of the three, however, has a clearly developed superficial hyphal mat surrounding the stromata. We consider that the wide concept of Rosellinia in the sense of Ju & Rogers is unacceptable, and prefer to extend the circumscription of Astrocystis by including species with such hyphal mats. We suspect that Rosellinia, even in the restricted sense of Petrini (1993) and Læssøe & Spooner (1994), is polyphyletic, and continuing study may allow the recognition of further segregate taxa. The terms ‘true’ and ‘false’ subiculum were adopted by Læssøe & Spooner after their introduction by Korf (1952) for use in the description of species of Arachnopeziza (Leotiales). The two types were distinguished by the relative positions of fruit- bodies and hyphae ; ascomata developing directly from a superficial hyphal mat were referred to as having true subicula, while ascomata which developed from the substratum but which were surrounded by superficial hyphae were considered to have false subicula. We feel that the application of this terminology to fungi unrelated to the Leotiales, such as the genera of Xylariaceae discussed here, needs further thought. We also prefer to simplify the definitions of technical terms as much as possible. These structures are referred to in the current paper simply as superficial hyphal mats. We unfortunately do not have access to cultures of these species, so information on anamorphs is not provided. Despite diligent searching, we must conclude that the small superficial hyphal mat surrounding the stromata of one described species (A. fimbriata) is sterile, and does not produce conidia. Mycol. Res. 102 (11) : 1325–1330 (1998) Printed in the United Kingdom Fungi from Mauritius : three Astrocystis species from Pandanus