Completeness of population-based cancer registration has been most commonly quantified by indirect measures, such as the death certificate only index or the mortality/incidence ratio. A major disadvantage of these measures is their strong dependence on the case fatality rate. Capture-recapture methodology offers an approach to estimate completeness directly which does not share this limitation. In this paper, a three-sources modelling approach is employed to derive estimates of completeness for the population-based cancer registry of Saarland. Overall, completeness is found to be high: estimates for all types of cancer range from 95.5 to 96.9% for calendar years 1970, 1975, 1980 and 1985. There is some variation with age (consistently high levels above age 30 years, a minimum of 87.7% in age group 15-29 years) and between cancer sites. Among the most common cancer sites, estimates of completeness are highest for gastrointestinal cancers (97.2%) and breast cancer (97.1%), while lower estimates of completeness are derived for cancers of the female genital organs (92.5%), the urinary tract (91.8%) and the prostate (91.0%). Although capture-recapture estimates are sensitive to the underlying assumptions about dependence between sources, careful application is encouraged to supplement traditional methods for evaluating completeness of cancer registration.