Debate regarding the putative ‘long peace’ since World War II has led to much quantitative analysis of historical battle deaths in search of changes in their statistical properties. It has been claimed and disputed that World War II marked the beginning of a decline of war on an unprecedented scale. We use changepoint analysis to analyse two well documented historical datasets. For heavy-tailed data such as the war deaths records, we test and calibrate techniques that make robust changepoint detection possible. We then demonstrate the existence of a number of distributional changes, regardless of data presentation. We conclude that there are changes around 1910 and 1950 CE, naturally bracketing the World Wars, along with weaker changes around 1880 and 1994 CE. Our analysis suggests a methodology for future investigations and provides a sound empirical basis for political and historical discussions of the severity and frequency of war.