Carcinogenic risk assessment involves a mixture of statistical, scientific, and public policy considerations. Concepts in current use, such as "no observed effect levels" and "virtual safety," and the problems in implementing them by means of dose-response models, particularly the probit-log dose and linear models are reviewed. The upper limits to risk provided by some conservative procedures are inconsistent with coherent balancing of risks and benefits. A common basis to the dose-response curves describing both carcinogenic and noncarcinogenic effects is to be found in deactivating reactions. A simplified model in which a toxic substance is activated and deactivated in separate and simultaneous reactions is presented and the dose response curve implied by the model is deduced. This curve has the general form of a hockey stick, with the striking part flat or nearly flat until the dose administered saturates the deactivation system, after which the probability of a response rises rapidly. Such a curve describes the Bryan-Shimkin methylcholanthrene-tumor incidence dose response curve as well as the probit log-dose model. The concept of a saturation dose is relevant to risk assessments for carcinogenic and noncarcinogenic substances alike.