The evolution of the diatoms has never been considered in any detail and this paper sets out some of the problems and discusses briefly the fossil evidence, the likelihood of monophyletic versus polyphyletic origin and the possible habitats of the ancestral forms. We assume that diatoms arose at a fairly early period and certainly long before the Cretaceous, when the first fossil diatoms were deposited. A `pre-diatom' stage is envisaged during which a naked photosynthetic cell acquired a coating of siliceous scales. This was followed by an `Ur-diatom' stage when the siliceous scales differentiated to form valves and girdle bands, to give a recognizable but extremely simple diatom. Evidence for such development is indirect and we discuss this from the standpoint of the structure of scales of modern diatom auxospores, symmetry of initial valves, homology of valves and girdle bands, comparative morphology of auxospore scales/valves/girdle bands and the developmental sequences of modern diatom valve formation. From a discussion of the available data and leaning heavily on electron microscope studies we suggest that there is very considerable evidence to substantiate our hypothesis. (a) It is proposed that at some time before the Cretaceous the spherical prediatom stage arose, possessing eukaryotic plant organelles and developed siliceous scales. (b) Such scales have been reported from the auxospores of several centric diatoms and their morphology is virtually identical to the simplest valve structure of centric diatoms. (c) The siliceous scales probably arose by specialization of cytoplasmic membrane systems associated with biochemical developments leading to silica deposition. We do not, however, exclude the possibility of a symbiotic event involving the incorporation of a silica `skeleton' consisting of plates, but we have no suggestion of any group of organisms that may have contributed such a wall. (d) The next stages involved the specialization of apical scales to form rudimentary valves. (e) We assume that girdle bands were also derived from scales, which then developed into hoop-like structures. Some modern genera, however, have scalelike cingula, which may represent a primitive state. (f) That both valves and girdle bands were derived from similar precursors is shown by the similarity of areolar plates in both components in many modern genera. The fimbriae on one edge of modern girdle bands are a relic of the fimbriate edges on the primoridal scales. Fimbriate edges on the bands around the auxospores of pennate diatoms are probably also relict features. (g) The existence of dome-shaped initial valves in centric genera that otherwise have Petri-dish-like or cylindrical vegetative valves, is further support for the hypothesis and may be a case of ontogeny providing a further indication of the phylogeny. (h) Developmental sequences of the valve of modern diatoms fit well with a scale origin, since both are formed from an initial central disk of silica from which silica ribs radiate, cross bars form and areolae are laid down in the spaces between costae. (i) Pore plates do not form in auxospore scales, which is further evidence that these are evolutionary young structures. (j) This hypothetical scheme of evolution is, we believe, well documented and can form an adequate basis for the derivation of the three classical series (centric, araphid and raphid) of diatoms.