Tropospheric NO2 vertical column densities (VCDs) over the Lombardy region were retrieved from measurements of the Global Ozone Monitoring Experiment (GOME) spectrometer for the period 1996--2002 using a differential optical absorption method. This data set was compared with in situ measurements of NO2 at around 100 ground stations in the Lombardy region, northern Italy. The tropospheric NO2 VCDs are reasonably well correlated with the near-surface measurements under cloud-free conditions. However, the slope of the tropospheric VCDs versus ground measurements is higher in autumn-winter than in spring-summer. This effect is clearly reduced when the peroxyacetyl nitrate and nitric acid (HNO3) interferences of conventional NOx analyzers are taken into account. For a more quantitative comparison, the NO2 ground measurements were scaled to tropospheric VCDs using a seasonal NO2 vertical profile over northern Italy calculated by the Model of Ozone and Related Tracers 2 (MOZART-2). The tropospheric VCDs retrieved from satellite and those determined from ground measurements agree well, with a correlation coefficient R = 0.78 and a slope close to 1 for slightly polluted stations. GOME cannot reproduce the high NO2 amounts over the most polluted stations, mainly because of the large spatial variability in the distribution of pollution within the GOME footprint. The yearly and weekly cycles of the tropospheric NO2 VCDs are similar for both data sets, with significantly lower values in the summer months and on Sundays, respectively. Considering the pollution level and high aerosol concentrations of this region, the agreement is very good. Furthermore, uncertainties in the ground-based measurements, including the extrapolation to NO2 VCDs, might be as important as those of the NO2 satellite retrieval itself.