Log-linear Poisson mixed models were used to study provincial malignant brain tumour (MBT) mortality among children and adolescents in Spain (1975-1992) in order to investigate the influence of specific socio-economic factors and to produce smoothed estimators of standardised mortality ratios (SMRs). Interdependence between geographical units was taken into account by including provinces as random effect terms nested with the corresponding Autonomous Region (Spain's administrative divisions). MBT mortality showed a positive association with three variables: non-cancer-related infant mortality; percentage of provincial land surface area devoted to agriculture; and industrial/construction activity. According to the final model, SMRs increased by 4% with every 10% rise in the area devoted to agriculture, with a 4% excess risk predicted for every increase in non-cancer-related infant mortality of 1 per 1000 person-years, and a 7% excess risk for contiguous categories of industrial/construction activity. By smoothing extreme values caused by random variability, the regression model yielded a reasonable estimation of SMRs. While infant mortality may be linked to the quality of medical care available, the relationship seen between MBT mortality and agricultural area accords with the excess risk reported for farmers' offspring. Finally, industrial activity might be regarded as a risk factor or as a marker of other conditions also associated with these tumours.