There have been significant gains in improving maternal mortality over the last two decades. Researchers have suggested a variety of interventions and mechanisms to explain these improvements. While it is likely that much of what has been done in research and programs has contributed to this decline, the evidence regarding what works in the settings in which women deliver continues to face many challenges. We review the evidence for these improvements and suggest that there remain areas to focus on, particularly the births which currently take place in an unsupervised or substandard environments. We highlight the main areas where more evidence is needed, and end with a call to determine which of our interventions seem to have the most benefit; which do not; and where to invest future resources.