The steepness of dose-response curves in radiation therapy depends to a large extent on the statistics of cell killing. This is so if the last few clonogenic tumor cells have to be hit or eradicated by other means to cure the patient. The steepness is dependent on the number of clonogenic cells in the tumor and the possible variation in their sensitivity. However, the uniformity of the dose distribution is also important and a decreased slope may result when the delivery of the dose is nonuniform or statistically uncertain. The variance in the energy imparted at the microdosimetric level to individual cell nuclei constitutes the ultimate limit of the variance in delivered dose at a given mean tumor dose. Considering all dosimetric variances it is shown that for low-LET beams the conventional microdosimetric variance will dominate, while in neutron and high-LET beams in general the microdosimetric variance may contribute significantly to the observed dose-response relationship. As a result the normalized slope of the dose-response curve for tumor control and normal tissue complications with neutrons and other high-LET beams will be reduced compared to that with photons. This conclusion is found to be in quantitative agreement with available data from clinical trials with neutron therapy. Finally, it is pointed out that for beams with a very high RBE and LET it may be favorable to deliver a fraction of the total dose in the form of conventional low-LET radiation. This addition of low-LET radiation may be desirable to ensure a dose to all clonogenic tumor cell nuclei that is sufficiently high and uniform to achieve a high probability of tumor control.