As part of a programme of landscape-scale habitat surveillance in the United Kingdom (UK), the effect of grassland sampling intensity on the outcome of numerical classification was assessed. Sample quadrats from two regions of the UK were available for post priori analysis; a random sample from Great Britain (GB), with grasslands sampled in proportion to area, and an independent stratified random sample from Northern Ireland (NI), with similar numbers of quadrats from agricultural and semi-natural grassland habitat strata. Classification of a combined area-proportional (balanced) random sample from GB and NI showed the species composition of UK grasslands to be determined largely by climate, landscape structure and land-use intensity. The classification was influenced primarily by the greater number of eutrophic agricultural grassland quadrats and semi-natural grassland quadrats of the larger GB study area. The semi-natural grasslands of NI, represented by a small number of quadrats, had little influence. Classification of a stratified NI sample combined with an area-proportional GB sample was influenced most by the NI semi-natural grassland quadrats. The structure of the classifications depended on sampling intensity. Vegetation classification should be derived from a balanced sample so that it is representative and its application does not lead to decisions being directed at classes of vegetation (or estimates derived from them) that are weighted by sampling intensity. Area-proportional sample design linked explicitly to landscape structure satisfies the requirement for a balanced classification. The issue of data-balance is relevant in conservation management and environmental assessment, where stratification is a commonly accepted procedure to reduce sampling effort, or is carried out to sample rare or ecologically interesting vegetation. It applies to landscape-scale vegetation classifications used for environmental assessments and to classifications that compare plant communities between regions (as in phytosociological studies). The issue is also important when combining environmental databases from international sources for classification purposes.