The cure rate of tumors was analysed by a model, and applied to human cancer. The cure rate increased linearly when plotted on a probit scale against dose in logarithm. The coefficient of variation, the ratio of one standard deviation to the dose of 50% cure, was taken as an indicator. The minimum value of the coefficient of variation was 5% obtained theoretically, and 10% clinically. This difference might be due to the variability existing in cancer of patients. The complications to the normal tissues, which was induced by the radiotherapy of cancer, were also analysed. The coefficient of variation for the complication showed the same tendency to that of the cure rate. The cures without complications were affected by the coefficients of variation for the cure rate and for the complication rate. The coefficient of variation for the cure rate is determined by the stochastic effect of cell killing and the error in dose estimation. When the physical error was 5%, its effect on the cure rate could be negligible. But when its error became more than 10%, a significant difference of the cure rate would be observed clinically in some tumors. The accuracy of the dose estimation was emphasized.