Comparisons are key to all fair tests of the effects of treatments. Sometimes patients experience responses to treatments which compare dramatically with past experiences and the natural history of health problems. In these circumstances, confident conclusions about treatment effects can be reached without carefully controlled research. Such dramatic effects of treatments are rare, however, and reliable detection of moderate but important differential effects of treatments requires carefully designed, formal comparisons. A key principle in such formal treatment comparisons is that like will be compared with like – that, before the treatment(s) to be assessed have been started, the patients in the treatment comparison groups should have similar chances of recovery. In the middle of the 20th century, random allocation to treatment comparison groups began to be adopted as an unbiased way of creating similar groups. It is widely assumed that the adoption of random allocation in controlled trials reflected the influence of RA Fisher’s development of statistical theory. The evidence suggests otherwise.