Standard methodology for analysis of air pollution epidemiological time series expresses effects in terms of relative risk, i.e. increases in the number of events associated with a short term increase in air pollution. However, even large relative mortality rates may in fact reflect a very small effect in terms of person-years life loss. In Zagreb, mortality in 1995–1997 was significantly associated with concentrations of nitrogen dioxide (NO2). We have used STL decomposition of time series into additive components of decreasing smoothness to test the hypothesis that mortality — air pollution association is due to short term mortality displacement. According to our results association between mortality and concentrations of NO2 remains statistically significant at time scales ranging from a few days to 1–2 months.