The various purposes of predictions of future cancer burden are outlined and divided principally into two contexts, administrative and scientific. Various difficulties in making predictions are reviewed, relating to unknown etiology of cancer, to changes in definitions, criteria and facilities, to weaknesses in the statistical base and to choice of the statistical model. Predictions made under the Nordic Action Plan against Cancer, particularly for Sweden, are given as an example. The other examples concern predictions of incidence of lung cancer and smoking in Europe and incidence of breast cancer and mammography screening in Finland.