There is much documentation in the scientific literature (e.g. Clay et al. 1999, Hamouz et al. 2006, Gerhards et al. 2012) that weeds are distributed in a patchy manner throughout agricultural fields. An aggregated distribution pattern of weed populations provides opportunity to reduce the herbicide application if site-specific weed management is adopted. Gerhards et al. (2002) achieved herbi-cide savings of 60% and 92% for dicotyledonous and monocotyledonous weeds, respectively, in spring barley (Hordeum vulgare L.) cultivation, and savings of 11% and 81% were achieved for the same weed groups for maize (Zea mays L.). In two winter wheat fields, Häusler et al. (1998) found that herbicide applications were necessary for 24% and 35% of the area for Galium aparine L., for 25% and 31% for other dicotyledonous weeds and for 55% and 7.5% for grasses. Using a simulation , Christensen and Heisel (1998) estimated a 40% reduction of herbicide consumption, and Nordmeyer and Häusler (2000) calculated the need for herbicide application from 7% to 64% of the total area in various fields. Site-specific weed management is based on the threshold concept: the aim is to adjust the intensity of management practices to the actual degree of weed infestation, with only those areas having a weed density exceeding the predefined control threshold typically being treated. However, site-specific weed management requires the precise setting of control thresholds for effectiveness and reliability. Furthermore, both areas that are treated superfluously and unsprayed areas that will show increases in weed infestation should be ABSTRACT An aggregated distribution pattern of weed populations provides opportunity to reduce the herbicide application if site-specific weed management is adopted. This work is focused on the practical testing of site-specific weed management in a winter wheat and the optimisation of the control thresholds. Patch spraying was applied to an experimental field in Central Bohemia. Total numbers of 512 application cells were arranged into 16 blocks, which allowed the randomisation of four treatments in four replications. Treatment 1 represented blanket spraying and the other treatments differed by the herbicide application thresholds. The weed infestation was estimated immediately before the post-emergence herbicide application. Treatment maps for every weed group were created based on the weed abundance data and relevant treatment thresholds. The herbicides were applied using a sprayer equipped with boom section control. The herbicide savings were calculated for every treatment and the differences in the grain yield between the treatments were tested using the analysis of variance. The site-specific applications provided herbicide savings ranging from 15.6% to 100% according to the herbicide and application threshold used. The differences in yield between the treatments were not statistically significant (P = 0.81). Thus, the yield was not lowered by site-specific weed management.