Agri-environment schemes have been implemented across the European Union in order to reverse declines in farmland biodiversity. To assess the impact of these schemes for bumblebees, accurate measures of their populations are required. Here, we compared bumblebee population estimates on 16 farms using three commonly used techniques: standardised line transects, coloured pan traps and molecular estimates of nest abundance. There was no significant correlation between the estimates obtained by the three techniques, suggesting that each technique captured a different aspect of local bumblebee population size and distribution in the landscape. Bumblebee abundance as observed on the transects was positively influenced by the number of flowers present on the transect. The number of bumblebees caught in pan traps was positively influenced by the density of flowers surrounding the trapping location and negatively influenced by wider landscape heterogeneity. Molecular estimates of the number of nests of Bombus terrestris and B. hortorum were positively associated with the proportion of the landscape covered in oilseed rape and field beans. Both direct survey techniques are strongly affected by floral abundance immediately around the survey site, potentially leading to misleading results if attempting to infer overall abundance in an area or on a farm. In contrast, whilst the molecular method suffers from an inability to detect sister pairs at low sample sizes, it appears to be unaffected by the abundance of forage and thus is the preferred survey technique.