The word "spectrum" gets tossed around a lot in mathematics, and there seem to be a number of different concepts to which it applies. There is of course a physical connotation to the word which is commonly associated with scattering processes, rainbows, etc. : http://en.wikipedia.org/wiki/Spectrum
However, in mathematics the term seems to be quite loaded. To name a just a few concepts which could be called spectral, here is a big list:
Eigenvalues/eigenvectors for normal matrices
Jordan normal form for matrices
Spectrum of a bounded linear operator
Spectrum of a C* algebra
Spectrum of a commutative ring / Scheme
Characters of an abelian group
Irreducible unitary representations of a group
Spectrum of a graph
Spectrum of a Riemannian manifold
Spectral sequences from cohomology theory
Of course some of these concepts are more general than others. For example, normal matrix < bounded linear operator < c* algebra < commutative ring. However, it doesn't seem like any one of these definitions is sufficiently encompassing to give the whole story. As a counter example, the irreducible representations of a non-commutative group are an instance of the Jordan normal form (for finite groups anyway), but are not really captured by the corresponding notion of the spectra of a commutative ring. Similarly, the spectrum of a graph and a Riemannian manifold don't seem to have much to do with the spectrum of schemes, but yet they are related to the spectra of linear operators. And then there are spectral sequences which are just a bit weird...
I won't profess to completely understand the general idea here, but there do seem to be some patterns. A common theme seems to be `decomposability', for example when finding the eigenvalues/vectors of a matrix one attempts to split apart the domain of the matrix into independent components. This is similar to splitting a space into points; and suggests that there is perhaps a correspondence between eigenvectors/eigenvalues and prime ideals/local rings. The non-commutative picture is of course more complicated, but perhaps the concept might be equally described in terms of irreducible modules and some type of generalized localization at an irreducible module (which is a hazy concept I must admit). In a perfect world, it would be nice if this same concept could even extend to things like the Gauss map from differential geometry, or the Legendre transform from statistical mechanics/convex optimization (of course that might not be feasible).
Also, I marked the question community wiki in case anyone else has some other good examples of `spectral' concepts in mathematics.