Survival from cancer over a certain time period is often measured by the 'relative survival rate'. This is the ratio of the observed survival rate in the group of patients to the survival rate expected in a group of people in the general population, who are similar to the patients with respect to all of the possible factors affecting survival at the beginning of the period, except for the disease under study. When patterns of patient withdrawal differ for a number of subgroups of patients with equal relative survival rates, the current method of derivation of the relative survival rate is biased. A method based on the concept of an 'expected life table' is proposed for removal of the bias. Examples based on material from the Finnish Cancer Registry suggest that the practical performance of the proposed method is better than that of other alternatives, even when the relative survival rates in the subgroups are not equal.