The selection pressure exerted by herbicides has led to the repeated evolution of herbicide resistance in weeds. The evolution of herbicide resistance on contemporary timescales in turn provides an outstanding opportunity to investigate key questions about the genetics of adaptation, in particular, the relative importance of adaptation from new mutations, standing genetic variation, or geographic spread of adaptive alleles through gene flow. Glyphosate-resistant Amaranthus tuberculatus poses one of the most significant threats to crop yields in the midwestern United States (1), with both agricultural populations and herbicide resistance only recently emerging in Canada (2, 3). To understand the evolutionary mechanisms driving the spread of resistance, we sequenced and assembled the A. tuberculatus genome and investigated the origins and population genomics of 163 resequenced glyphosate-resistant and susceptible individuals from Canada and the USA. In Canada, we discovered multiple modes of convergent evolution: in one locality, resistance appears to have evolved through introductions of preadapted US genotypes, while in another, there is evidence for the independent evolution of resistance on genomic backgrounds that are historically nonagricultural. Moreover, resistance on these local, non-agricultural backgrounds appears to have occurred predominantly through the partial sweep of a single haplotype. In contrast, resistant haplotypes arising from the midwestern US show multiple amplification haplotypes segregating both between and within populations. Therefore, while the remarkable species-wide diversity of A. tuberculatus has facilitated geographic parallel adaptation of glyphosate resistance, more recently established agricultural populations are limited to adaptation in a more mutation-limited framework.