Overdiagnosis is defined as the detection by screening of a cancer not destined to present clinically in the person's lifetime. It is inevitable when screening for cancer, and becomes more prevalent the more sensitive the screening test is in detecting very early pre-clinical cancers. It has been demonstrated for lung1 and prostate cancer screening2, and is also probable for colorectal screening, though the overdiagnosed cancers are masked by the reduction in colorectal cancer incidence caused by the detection and treatment of colorectal cancer precursors. The same is almost certainly true for screening for cancer of the cervix. Controversy concerning overdiagnosis for breast cancer has arisen over differences in the estimates that have been made in the extent of overdiagnosis3, 4, and this controversy is reflected in the three letters we publish in the current issue of the IJC. The letters all concern an analysis of data from the Breast Cancer Screening Program of Norway5, and they largely relate to differences over methodological issues. However, all the authors agree that the method used by Falk et al5 to estimate the extent of overdiagnosis was not optimal. The best method is derived from a randomized screening trial, when subjects in the screening arm receive screening for a defined period, then the screened and control groups are followed for a sufficiently long time without screening for the effects of lead time from screening to dissipate. The difference in cumulative numbers of cancers is then equivalent to the degree of overdiagnosis, that difference is expressed as a proportion of the screen-detected cancers. Such a method cannot be applied in a screening trial where the controls are offered screening at the end of screening in the intervention group, hence overdiagnosis could not be validly measured in the Swedish Two-county trial, even though an attempt was made to do so6, but it could in the Malmö7 and Canadian breast screening trials8. For the Malmö trial the estimate of overdiagnosis was 10%. For the Canadian trial, an update is expected soon. The estimates of overdiagnosis that have been made using population data vary from very little, to 50% or more9. The estimates of Falk et al5 fall about the middle of this distribution. Many choose to discount these estimates, as it is difficult to account for lead time, but they are important, as they help to place the possible extent of overdiagnosis in perspective with the likely benefit from the screening that resulted in overdiagnosis, as in a recent US estimate of 30%, compared to a reduction of advanced disease of only 8%10. Overdiagnosis confounds the assumption of benefit from screening that many make, especially the public, screening advocates, and regrettably many physicians, who equate the early detection of cancer with benefit. Yet this is a false assumption11. Screening remains an expensive use of resources; as specialists in imaging and biomarkers develop more and more sensitive screening tests, the extent of overdiagnosis is bound to increase, and we have to remain aware of this to avoid assumptions of increasing benefit when none has been demonstrated. Anthony B. Miller Dalla Lana School of Public Health, University of Toronto