Abstract The rehabilitation of native plant communities in urban bushland remnants is an increasingly important activity requiring the collection of large amounts of seed. Best practice generally identifies that local seed are best, but how far does the local provenance extend? Using the DNA fingerprinting technique amplified fragment length polymorphism, we assessed genetic differentiation between potential seed source populations and the target population, Bold Park, a large and significant bushland remnant in Perth, Western Australia. For each of 15 species, analysis of molecular variance was used to partition genetic variation within and among populations. Genetic differentiation between Bold Park and potential seed source populations was assessed by non-metric multidimensional scaling ordination, and statistically by Fisher’s exact tests. The partitioning of variation among populations (ΦST) varied from 0.66 for Santalum acuminatum to 0.04 for Mesomelaena pseudostygia. For eight of 15 species, Bold Park plants were completely or largely non-overlapping with other populations in ordinations, suggesting genetic differentiation and a narrow provenance. Five species showed overlap between Bold Park and some other, but not all, populations sampled, with geographically closest populations generally undifferentiated. Only two species, Acanthocarpus preissii and Mesomeleana pseudostygia, showed little genetic differentiation between Bold Park and all other populations, suggesting a regional genetic provenance. These species can be classified into three broad provenance classes – narrow, local and regional – to help guide decisions about appropriate seed-collection zones for the rehabilitation of urban bushland remnants.