Epidemiological studies provide the primary data on the carcinogenic effects of radiation in humans. Much of what is known has come from studies of the atomic bomb survivors, and to a lesser extent from patients receiving radiotherapy. These studies demonstrate that exposure to moderate to high doses of radiation increases the risk of cancer in most organs. For all solid cancers combined, cancers of the thyroid, breast and lung, and leukemia, risk estimates are fairly precise, and associations have been found at relatively low doses (<0.2 Gy). Associations between radiation and cancers of the salivary glands, stomach, colon, bladder, ovary, central nervous system and skin have also been reported, but the relationships are not as well quantified. Associations between radiation and cancers of the liver and esophagus, and to a lesser extent multiple myeloma and non-Hodgkin's lymphoma, have been reported in a few studies, but results are inconsistent. Chronic lymphocytic leukemia, Hodgkin's disease, and cancers of the pancreas, prostate, testis and cervix have rarely been linked to radiation exposure. A linear no-threshold model adequately describes the dose-response relationship for solid cancers, although at extremely high doses the risk appears to flatten out. Because few populations have been followed until the end of life, the temporal patterns of risk are not completely known. An increased risk, however, does continue for several decades. In contrast, radiation-related leukemias begin to occur shortly (2-3 years) after exposure and, at least in the A-bomb survivors, a linear-quadratic dose response seems to fit the data better than a pure linear model. Radiation does not act entirely in isolation. It can interact with other carcinogens, e.g. tobacco or chemotherapeutic agents, and with host factors such as age at exposure, gender or reproductive history. Interactions with medical interventions or with certain heritable mutations have also been suggested. While the studies of high-dose exposures are essential for understanding the overall biological consequences of radiation exposure, the public is more concerned about the long-term health effects from protracted exposures at low doses. Unfortunately, the inherent limitations of epidemiology make it extremely difficult to directly quantify health risks from these exposures. While most epidemiological data are compatible with linear extrapolations from exposures at high doses or high dose rates, they cannot entirely exclude other possibilities. As the field of epidemiology advances, understanding more about the health effects of prolonged and low-dose exposures will be the next challenge.