Here's a more mathy way of understanding it. You know how Taylor series work in Calculus? You have some differentiable function and you are able to approximat

Here's a more mathy way of understanding it. You know how Taylor series work in Calculus? You have some differentiable function and you are able to approximate the function arbitrarily well by adding up more and more polynomials. Turns out you can do the same thing with linear combinations of sines and cosines, with some slightly different conditions on the function. In particular, we need the function to be periodic. This is all that's happening: by adding up sines and cosines so they cancel out "just so," we can approximate other functions. The Fourier transform is the tool by which we calculate the coefficients of the various sines and cosines in the linear combination. Now, if we relax the notion that we require the function to be periodic, we get the "integral" form of the Fourier transform. We do this by slowly extending the permitted period to infinity since any function is periodic over a closed interval, although perhaps with a period of 1. This shouldn't be too surprising as the integral is a pretty specific sense the "continuous" version of a sum. When people talk about "frequency domain" what they mean is the Fourier transform "at x" is the contribution of "frequency x" to the overall value of the function. I give you the frequency, you give me its contribution. In fact, there's a very general theorem called the Stone-Weierstraß Theorem which tells us when you can approximate a function with linear combinations from a set of other functions. You can then go through a similar process to determine the extent to which each of these other functions contribute to the value of the original. This is a lot of hand waving and will most likely upset any mathematicians in the audience. I left out mountains of details. :) If you're comfortable with the idea of vector spaces and orthonormal bases, then the existence of the Fourier series is equivalent to the statement that linear combinations of sines and cosines have a dense span in the space of functions whose integral over the entire real line is finite. Yes, our vectors are functions and the sines and cosines don't quite form a basis, but they do form a "dense span," which is another way of saying that for a given function there's some linear combination that can approximate our functions up to any level of precision we desire. Here's another statement that wraps up Fourier transforms, Fourier series, and all that in one go. Let C([0,1]/{0,1}) be the space of functions continuous on [0,1] where we identify 0 and 1 as the same point to obtain a circle. Consider the family of functions f_n(x) = e^{2πinx} where n is an integer. Then the set {f_n : n is an integer} has a dense linear span in C([0,1]/{0,1}). One consequence of this is that the f_n form an orthonormal basis for L^2([0,1]), the space of "square-integrable functions" on [0,1] (http://en.wikipedia.org/wiki/Square-integrable_function http://en.wikipedia.org/wiki/Square-integrable_function). Remember that e^{ix} = cos(x) + isin(x), so f_n(x) = e^{2πinx} = cos(2πnx) + isin(2πnx), which is where the sines and cosines come from. Sorry for the fact that "pi" and "n" look so similar. It's HN's font. Anyhow, the above may or may not have been useful, I just thought I'd throw in a more mathy explanation than one usually sees. It didn't really click for me until I was able to relate what was happening with the Fourier series and Fourier transform to how I understood vector spaces and bases.