Analysis of air traffic control capacity shows the relation between traffic level and risk can be dealt with quantitatively only if units of risk are specified. Measuring risk in terms of fatalities per hour of exposure solves the major problem of how to compare the effects that many different kinds of risk have upon patterns of human activity, but leaves us with a number of unresolved secondary issues. Among these are : incommensurability of certain risks, failure of statistical independence, certain limitations of time-normalization, unequal distribution of costs and benefits, different weighting of voluntary and involuntary risks, liability, cost and value of uncertainty, positive value of some risks, acceptability of unequal risks, difficulty of estimating very small probabilities and threat of the unknown compared to threat of the known. Although these issues arose out of air traffic control studies, they are clearly relevant to risk analyses of many other systems.