[en] Complete text of publication follows. Objective: As the Linear Non Threshold model was first established in 1930, based on the results of a genetic study using Drosophila, we tried to indicate the conditions under which LNT holds. Methods: A sex-linked recessive lethal mutation assay was performed in Drosophila melanogaster. DNA repair proficient immature spermatocytes and spermatogonia were irradiated with X-rays at a high or low dose-rate. Results and discussion: Mutation frequency in the sperms irradiated with a low total dose (0.2Gy) at a low dose-rate (0.05Gy/min) was significantly lower than that in the sham-irradiated group whereas irradiation with a high dose (10Gy) at the same dose-rate resulted in a significant increase in the mutation frequency. It was obvious that the dose response relationship was not linear, but U-shaped. A low dose irradiation at a high dose-rate (0.5Gy/min) did not cause a significant reduction in mutation frequency. Mutation in the high dose, high dose-rate group was more frequent than in the high dose, low dose-rate group. A dose-rate effect was evident. When mutant male flies defective in DNA excision repair function were used instead of wild type flies, a low dose irradiation at a low dose-rate did not cause the reduction in the mutation frequency. These observations suggest that the dose response relationship is dependent not only on the dose-rate, but also on the DNA repair function. It is inferred that error-free DNA repair functions were activated by a low dose of low dose-rate irradiation, and this repaired spontaneous DNA damage rather than the X-ray induced one, thus forming a practical threshold. As the human somatic cells are usually repair function proficient, we conclude that LNT can not estimate the human cancer risks properly.