OBJECTIVE: This Belgian study assessed the geographical and temporal differences in the exposure of the population to inorganic arsenic, a known carcinogen. METHODS: In the CadmiBel study (1985-9) the 24 h urinary arsenic excretion was measured, as an index of recent exposure, in industrialised cities (Liège: n = 664, Charleroi: n = 291), in a rural control area (Hechtel-Eksel: n = 397), and in rural districts in which the population had possibly been exposed through the drinking water or the emissions of nonferrous smelters (Wezel: n = 93, Lommel: n = 111, and Pelt: n = 133). In the PheeCad study, in 1991-5, the rural areas (n = 609) were re-examined together with an urban control area (Leuven: n = 152). RESULTS: The CadmiBel results showed that after adjustment for sex, age, and body mass index, the 24 h arsenic excretion was on average low in Liège (91 nmol), Charleroi (155 nmol), Hechtel-Eksel (144 nmol), and Wezel (158 nmol), whereas the highest excretions were found in Lommel (570 nmol) and Pelt (373 nmol). During the PheeCad study, the mean 24 h arsenic excretion in the rural areas ranged from 81 to 111 nmol. This was lower than six years earlier and similar to the excretion in the control town (108 nmol). Longitudinal studies in 529 people living in the rural areas confirmed that their 24 h arsenic excretion had decreased (P < 0.001) from 222 to 100 nmol. As well as the drinking water, industry was likely to be a source of the increased exposure in Lommel and Pelt in 1985-9, because at that time the urinary arsenic excretion did not follow the regional differences in the arsenic content of the drinking water, because the fall in the arsenic excretion over time coincided with the implementation by industry of stricter environmental regulations, because in individual subjects the urinary arsenic excretion was inversely correlated with the distance to the nearest smelter, and because an increased arsenic excretion was only found downwind from the main smelter. The official network that monitors the arsenic concentration in airborne and fall out dust did not detect the high exposure in Lommel and Pelt between 1985 and 1989. CONCLUSION: This study highlights the necessity to validate environmental monitoring programmes by directly estimating the internal exposure of the population.