Regression towards the mean was demonstrated in semen analyses from 216 donors for artificial insemination by selecting those with low sperm concentration (less than 40 X 10(6)/ml), low motility (less than 50%) or low motility index (less than 120) in the first test. For each characteristic, the result was significantly higher in the second test, the means for first and second tests being, for sperm concentration 28 and 56 million per ml (n = 14, P less than 0.01), for motility 42 and 55% (n = 14, P less than 0.01) and for the motility index 95 and 131 (n = 24, P less than 0.001). Regression towards the mean occurs whenever extremes of the range of a variable and selected for re-measurement. Many of the current treatments for oligospermia and asthenospermia have been promoted because semen quality improved during uncontrolled studies in which this phenomenon may have been operative. Using means of multiple samples before and following therapeutic intervention will not prevent regression towards the mean, only reduce its magnitude. Thus, placebos are essential in the evaluation of treatments for male infertility.