Considerable increase in cancer deaths has been observed since 1995, after when the tenth version of the international classification of disease (ICD) was introduced in Japan according with the revision of a death certificate form at the same time. We assessed the contributing factors for this unnatural fluctuation, using a population-based cancer registry data as a model. All deaths of the prefectural residents are collated with the cancer registry database in the registration process. For all Japanese deaths of Aichi Prefecture in 1994 (n=41,111) and in 1995 (n=42,944), the description of the death certificates were compared with the ICD classified as the cause of death. It was ascertained that 97-99% of cancer and 97-98% of non-cancer deaths classified by ICD were proper. Among those classified as cancer as the cause of death and stated as other than the direct cause of death (n=92 in 1994, n=428 in 1995), only 22% (1994) and 8% (1995) were considered to be classifiable to both cancer and non-cancer as the cause of death. Among those who were classified as non-cancer as the cause of death in spite of including cancer in the original statement (n=770 in 1994, n=638 in 1995), 24% (1994) 10% (1995) were considered to be non-cancer as the cause of death, and the rest were considered classifiable into both cancer and non-cancer. Even if all cancer deaths classifiable to non-cancer were classified as non-cancer deaths, change in the number of cancer deaths was inconsiderable in both years. Therefore, it is suggested that the increase of cancer deaths after introduction of the new ICD was caused not by the artifact such as change in selection rule of disease when classify the cause of death, but by the improvement of the description of death certificate by physicians as a result of revision of death certificate form.