Understanding the role of unmanaged arthropod flower visitors as crop pollinators is critical if robust and reliable long-term alternatives are to be found for honey bee pollination. However, data on pollinator assemblages can be scant. Field observation of crop flower visitors is a common data collection technique but it can be inadequate for species identification and is labour-intensive if used across many sites. Trapping may reduce this problem, but trap performance and sampling consistency over long distances (sites separated by >100 km) are rarely examined. Window traps were designed to collect flower-visiting arthropods from peak-flowering onion (Allium cepa) and pak choi (Brassica rapa var. chinensis) fields across several regions throughout New Zealand. Trap efficacy was evaluated by comparing trapped samples with observations of flower visiting arthropods during the same trapping period, from dawn (6:00 to 7:00 hours) through to dusk (20:00–21:00 hours) at the same locations. Similar types of larger arthropods (length ≥3 mm) were observed and trapped within both crops, with the hymenopteran genera Apidae, Colletidae and Halictidae and the dipteran families Syrphidae, Calliphoridae, Anthomyiidae, Stratiomyidae, Sarcophagidae, Bibionidae, Tachinidae and Muscidae the most commonly recorded. The total counts of these taxa across fields were strongly correlated between the two methods; however, the ratio of trapped to observed individuals could vary greatly between taxa. Trapping allowed more arthropods to be identified to the species level and also helped record more small arthropods (body length <3 mm) when compared with observation. Window traps can be effective for assessing the relative diversity of flower visitor assemblages and the abundance of specific taxa in specific cropping systems at the regional scale, but variation in trap efficiency between arthropod taxa must be assessed for a true measure of assemblage composition.