Mortality is probably the earliest health outcome to be studied using statistical methods. While it was recognized for many centuries that mortality increased in cities during periods of air stagnation associated with air pollution, it was only in the last century that this was accepted scientifically as due in part to air pollution rather than to the cold weather which accompanies temperature inversions. Air pollution epidemiology has advanced by studying both temporal and spatial variations in pollution and outcome. Methods of investigation have improved due to developments in measurement of exposure, data processing and statistical theory but confounding is a persisting and possibly inherent problem. There have also been major changes in mechanistic and causal concepts which have been less obvious but which have important implications. This paper suggests that the most important development in mechanistic theory is that pollution, particles in particular, have the potential to cause systemic and not just respiratory effects. On the causal side, the most important development has been the transition from a single cause with threshold to a multifactorial causal model. This has important implications for understanding epidemiological results which show effects at low exposure with little evidence of a threshold. The latter property has profound implications for public health policy as it directs abatement strategies to a population exposure reduction approach rather than a limit or “hot spot” control approach.