Thanks, Jonathan. The reference contains a brief description of terms related to seed anatomy and morphology (seed shape, color and texture). This may give a basis for classification but is not in itself one.
I´m not an especialist, but Baskin & Baskin (2001) is a really important book in seed ecology (the title is "Seeds: Ecology, Biogeography, and Evolution of Dormancy and Germination") and I´m pretty sure they will provide an ecologically relevant classification of seeds.
Thanks, Santiago. Yes. There is a classification in pag. 28 of Baskin and Baskin. It is the classification by Martin (1946) in 12 seed types based on embryo morphology, relative amount of endosperm and position of the embryo. The types are: Broad, capitate, lateral, peripheral, rudimentary, dwarf, micro, linear, spatulate, investing, bent and folded. The authors give almost no explanation of each type. The figures illustrating each type (included in Table 3.2) are small drawings based exclusively in embryo size and position. Differences between types are not clear (for example between rudimentary, dwarf and micro the only difference consists in relative size of seed and embryo; micro seems to be a group reserved to Orchidaceae).
A conclusion from this classification could be that seed type is a conserved property of families, but I even doubt the generality of this. For example, Fabaceae are mentioned as having bent seeds. To confirm this we need to know better what are bent seeds and this is not explained.
Have you tried looking at the work of angiosperm palaeobotanists such as Margaret Collinson (e.g. Collinson, M. E. 1983. Fossil plants of the London Clay. Palaeontological Assiociation, London (Field Guide to Fossils), 1-121)? Fossil floras such as from the London Clay consist predominantly of angiosperm seeds and taxonomies for them have been developed that rely exclusively on seed morphology and anatomy. I am not suggesting you use their taxonomies, but their approaches to classifying seeds might be helpful.