The blast and resultant firestorm from the atomic bomb dropped on Hiroshima on 6 August 1945 are estimated to have killed some 70,000-80,000 people straight away. Some three quarters of these were civilians. Another 70,000 people were injured. The total dead has been estimated at between 90,000 and 140,000. Nearly two-thirds of buildings in Hiroshima were completely destroyed.
Civilian casualty ratio
Estimates vary, but a comprehensive examination of the data suggests that civilians have accounted for around half of all war deaths from century to century. For a visual representation, see Max Roser’s chart at Our World in Data (Figure 1, below).
Figure 1. Global Deaths in Conflicts Since 1400 (total [civilian plus military] 15 year moving average deaths per 100,000 world population shown by the red line, military deaths [also per 100k] in blue)
The ratio may be changing. In the 1990s, 90% of deaths from conflicts were civilians. 
World War II civilian casualties
World War II is widely accepted as the deadliest military conflict in human history. Figures vary considerably, but a generally accepted estimate is that some 75 million people (3% of the world’s population) died, of whom up to 55 million were civilians. The civilian to combatant fatality ratio in World War II is approximately 2:1, or 67%. Civilian deaths in the Soviet Union alone are put at 19 million. A quarter of the people in the Soviet Union were wounded or killed.
Apart from deaths and injuries, much suffering is caused by indirect effects: internal (within country) and external (between country) displacements; malnutrition and/or starvation; and disruption or ablation of normal health care arrangements are all common. We have seen that only too well in conflicts over recent years: Bosnia; Rwanda; Iraq; Afghanistan; Syria. For example, the UN Office for the Coordination of Humanitarian Affairs (OCHA) projected that 11.7 million people in Syria would require humanitarian and protection assistance in 2019.
Since 2011, the war in Syria has claimed thousands of lives and left 11 million people in need of humanitarian assistance. More than half the country’s population has been forced to flee their homes, including 6.6 million refugees seeking safety in neighbouring countries and another 6.7 million who are internally displaced inside Syria
The infrastructure needed to record the data accurately are often early casualties of war. Everything focuses on the crisis at hand rather than monitoring the effects of it—often written off as ‘collateral damage’. In their book, Seybolt et al.  examine the difficulties of even just counting civilian deaths, never mind quantifying the less tangible aspects of conflict (psychological impact; disease; malnutrition; social, economic, cultural and intellectual disruption; etc).
Thus, all published statistics can only be estimates, since the impact of political, and other, pressures from interested parties ensure that any given figure will be challenged by some party with a vested interest in inflating or minimizing the number—or discrediting the source—for their own purposes.
That said, modern methods of information acquisition, storage, retrieval, collation, analysis and presentation mean that we can have a better understanding then previous generations, although there are the potential problems of information overload, and ‘fake news’.
Whom to believe?
One can only cite one’s sources, aiming for the most objective, academically and methodologically robust, and dispassionate information. Some examples include (in alphabetical order):