Estimates of values in engineering, especially statistical estimates, are often given without concurrent estimates of confidence in those values, unlike those in the medical sciences. Values of probabilities and severities which form the axes of a risk matrix are given in terms of four or five discrete categories. This can be done for confidence values also. I present a calculus of discrete categories of confidence which allow confidence to be compounded, so that the confidence in, say, arithmetical functions of estimates can be derived.