Abstract Palmes diffusion tubes have been used to measure nitrogen dioxide concentration at 363 urban sites throughout the United Kingdom during the period of July–December 1991. Average concentrations over the period ranged from les than 10 ppb in northern Scotland to around 50 ppb at near-road sites in London. A total of 243 sites provided valid data for both this survey and an earlier, similar, study in 1986. On average, concentrations were about 34% larger in 1991. Increases occurred throughout the country and were not confined to any particular area or region, although the percentage change tended to be larger in the north and west and where concentrations were small. The observed difference in concentrations was consistent with differences in meteorological conditions between the periods covered by the two surveys and the 38% increase in emissions of NO x from motor vehicles over the period. However, there is no evidence from continuous monitoring between 1987 and 1991 of a marked trend in concentration at any one site. The shortening of the diffusion path in the diffusion tube due to wind effects has been demonstrated. This leads to a tendency for diffusion tubes to overread relative to chemiluminescent analysers. However, if the tubes are mounted in a sheltered location the overestimate is small. Since, in this survey, most of the samplers were mounted close to the sides of buildings, the data were not corrected, although the concentrations may be overestimates at some sites. In order to provide information on the spatial distribution of NO 2 over the whole country, population density (related to vehicle density) was used along with the survey results together with additional data on rural concentrations, to map NO 2 concentrations over Great Britain. This suggests that around one third of the population lives in regions where the mean concentration exceeds the European Community Directive Guide Value for median concentration. However this proportion would be reduced if the tendency for the tubes to overread and the fact that the mean value is generally larger than the median were taken into account.