This paper examines the survival of elderly European cancer patients, on the basis of the EUROCARE II results. Using Hakulinen and Abeywickrama's method, the relative survival rates at 1 and 5 years from diagnosis were computed by sex and quinquennial age group for the elderly (65-99 years old). Age-standardised rates for the whole elderly group were also calculated. The analysis covered: all malignancies combined, stomach, colon, rectum, pancreas, lung, melanoma, bladder, kidney and non-Hodgkin's lymphomas for both sexes; prostate and larynx for men; and breast, ovary, uterine cervix and corpus for women. Data relating to 701521 cancer patients came from 44 population-based cancer registries in 16 European countries. The relative risks of death (RRs) of older patients (65-99) with respect to middle-aged adults (55-64) were computed by sex and country, for all malignancies only. The most prominent finding was the decrease in survival rates with increasing age for almost all cancer sites. The age-curves of survival rates at 1 year from diagnosis usually had a steeper slope than those at 5 years, particularly in women. This suggests that disease stage at presentation plays an important role in determining survival, particularly in the elderly. Thus, all factors which influence timing diagnosis in the elderly and cause a delay in tumour detection, such as psycho-social factors, access to care, co-morbidities and other clinical features affecting performance status, are very important predictors of prognosis. Very large geographic variations in relative survival rates were found among European countries. The ordering of countries was similar for almost all cancer sites. Western and Central Europe generally had the best survival, followed by Northern countries and by Southern ones (the latter with survival around the European average: 39% in men, 47% in women). The UK had survival rates unexpectedly lower than rates of nearest nations, often below the European average. Eastern countries usually had the lowest rates. In the very elderly patients (over 85 years), an apparent rise in the survival rates was noted, particularly at 5 years from diagnosis and in men. This 'too good' survival is unlikely to be due to real better prognosis, but rather to a selection bias. Countries with this unusual rise are also those registering a high proportion of DCO cases (those cases retrieved by death certificate only) (around 10%) or DCO unavailable. Another 'natural' bias has also to be taken into account: in elderly patients with a very bad prognosis, who are often suffering from other serious co-morbid conditions, cancer diagnoses could be under-notified and not reach at all the data sources commonly monitored by cancer registries.