A general form of ordinal association is defined in a prediction analysis framework. This definition provides a useful way of identifying and selecting prediction rules a priori. Various classes of association are considered and a general approach is provided for the selection and significance testing of optimal prediction rules (ex post). This methodology is also extended to the degree‐k, k ≥ 2, and multivariate cases. Nearly all of the common measures of ordinal association are special cases of the general type of association considered here.