This paper examines the effect of variable reporting and coding practices on the measurement of maternal mortality in urban and rural Scotland, 1861-1901, using recorded causes of death and women who died within six weeks of childbirth. This setting provides data (n = 604 maternal deaths) to compare maternal mortality identified by cause of death with maternal mortality identified by record linkage and to contrast urban and rural settings with different certification practices. We find that underreporting was most significant for indirect causes, and that indirect causes accounted for a high proportion of maternal mortality where the infectious disease load was high. However, distinguishing between indirect and direct maternal mortality can be problematic even where cause of death reporting appears accurate. Paradoxically, underreporting of maternal deaths was higher in urban areas where deaths were routinely certified by doctors, and we argue that where there are significant differences in medical provision and reported deaths, differences in maternal mortality may reflect certification practices as much as true differences. Better health services might therefore give the impression that maternal mortality was lower than it actually was. We end with reflections on the interpretation of maternal mortality statistics and implications for the concept of the obstetric transition.