The Bacillariaceae is a very species-rich family of raphid diatoms and includes the large and taxonomically difficult genus Nitzschia, whose species are often small-celled and finely structured and have few discrete morphological characters visible in the light microscope. The classification of Nitzschia is still mostly based on one developed in the second half of the 19th century by Grunow, who separated the genus into a series of sections largely on cell shape and symmetry, the position of the raphe, transverse extension of the fibulae, and folding of the valve. We assembled and analysed single-gene and concatenated alignments of nSSU, nLSU, rbcL, psbC and cox1 to test Grunow?s and subsequent classifications and to examine selected morphological characters for their potential to help define monophyletic groups. The maximum likelihood trees were equivocal as to monophyly of the family itself but showed good support for each of eight main clades of Bacillariaceae, three of which corresponded more or less to existing genera (Hantzschia, Cylindrotheca and Bacillaria). The other five main clades and some subclades comprised groups of Nitzschia species or assemblies of Nitzschia species with other genera (Pseudo-nitzschia, Fragilariopsis, Neodenticula, Tryblionella, Psammodictyon). Relationships between most of the eight main clades were not resolved robustly but all analyses recovered Nitzschia as non-monophyletic. The Grunowian classification of Nitzschia into sections was not supported, though in some respects (e.g. treatment of sigmoid species) it is better than subsequent reclassifications. Several of the main clades and subclades are cryptic (lacking morphological synapomorphies) and homoplasy is common in both light microscopical and ultrastructural characters (to the extent that organisms initially assigned to the same species sometimes prove to belong to a different main clade). Nevertheless, some characters, including the structure of the raphe canal and girdle, seem to be sufficiently conservative evolutionarily to give a provisional estimate of relationships if mo-lecular data are unavailable. No new formal classifications are proposed but various options are explored and research needs identified.