When Enola Gay pilot Paul Tibbets died November 1, the New York Times repeated Tibbets's contention that "It would have been morally wrong if we'd have had [the atomic bomb] and not used it and let a million more people die." That virtually no reputable historian would put the casualty figure for a US invasion of Japan anywhere near that high (leaving aside the question of whether an invasion would have been necessary) was not mentioned in the story.
One of my regular correspondents, the military historian D.M. Giangreco, wrote a definitive paper on the administration's casualty estimates, published as "'A Score of Bloody Okinawas and Iwo Jimas': President Truman and Casualty Estimates for the Invasion of Japan", in Pacific Historical Review, Feb 2003, and reprinted in Hiroshima in History: The Myths of Revisionism, ed. Robert J. Maddox, 2007, pp. 76-115. From his scrutiny of primary sources, he observed: "Truman's much-derided accounts of massive casualties projected for the two-phase invasion of Japan are richly supported by US Army, White House, Selective Service and War Department documents produced before the use of nuclear weapons against Japan and stretching back through the last nine months of the Roosevelt administration."
The American casualty figure was over 25,000 at Iwo Jima -- a piece of rock barely over 20 square kilometers. Out of the 21,000 Japanese soldiers who defended that island, less than 300 survived. On the basis of such heavy losses and fanatical Japanese resistance, it is reasonable to predict that American casualties would have approached the million mark upon, and leading to, the invasion of mainland Japan.