Marital differentials in survival from 12 common types of cancer are assessed by estimating a mixed additive multiplicative hazard regression model on the basis of individual register and census data for the whole Norwegian population. These data cover the period 1960-91 and include more than 100,000 cancer deaths. The data and method make it possible to take into account the marital mortality differentialsin the absence of cancer. The excess all-cause mortality among cancer patients compared with similar persons without a cancer diagnosis is, on the whole, more than 15% higher for never-married men, never-married women and divorced men, than for the married of the same sex. Other previously married have an excess mortality elevated by about 7%. This protective effect of marriage is not due to stage, which is controlled for. The possible importance of treatment and host factors is discussed.