The central idea of the business cycle is of a pervasive cyclical movement of economic indicators. This paper shows that the concept of such simultaneous movements can be given a precise meaning by performing a principal component analysis of spectral density matrices or, with a different shade of meaning, coherence matrices. It suggests also a new method of computing spectral approximations for models that are nonlinear in their variables. The methods are applied to the Klein-Goldberger model for the United States.