Agricultural set-aside and compensation land provide restoration ecologists with opportunities to re-create semi-natural habitats. Restoration sites often have high soil fertility and inadequate seed banks of desirable species. Sowing additional seed is a proven method for establishing chalk grassland vegetation. If seed of local provenance is required, it may be collected by hand or by using specialized machines. Ultimately, the mix collected must provide seed suitable for recreating vegetation similar to that of the donor site. We examine the ability of a vacuum machine to meet this requirement by comparison with hand collection, and we discuss possible effects on invertebrates. Microscope analysis and glasshouse and field trials were used to compare the abundance of seed of different species in harvested mixes with the vegetation composition of the donor site. Seed heads of individual species were examined to determine the number of viable seeds per head and attack rates by phytophagous insects. The mix contained seed of over half the species recorded on the donor site. The seed of taller, more common species was overrepresented in the mix, at the expense of some smaller, mat-forming plants. After one season, however, the vegetation of the field trial plots was of the same type as that of the donor site, although the proportions of the constituent species differed slightly and certain species were absent. Mechanical collection is more efficient than hand collection. Endophagous invertebrates are unlikely to be affected by the machine. Seed collection requires a combination of methods, precise timing, and careful planning to provide a full range of species and to minimize impacts on plant and invertebrate populations.