Prevalence data provided by cancer registries are generally biased, since the patients that were diagnosed before the starting of the registry's activity cannot be included in the statistics. The relevance of this incompleteness bias is estimated in this paper. Incidence and relative survival are modelled as parametric functions describing a wide class of cancer diseases. Prevalence estimates are then computed considering different hypotheses on disease reversibility. The ratio between the prevalence observed by the registry and the total estimated prevalence is used as an index of completeness. An analytical evaluation of this ratio, as a function of the parameters characterizing the observational process and the biological behaviour of the disease, is given.