(Statistics, Statistical Mechanics):
A microstate is another of those convenient fictions physicists use to give their work some degree of mathematical rigor. In statistical mechanics a microstate is a particular configuration of the particles in a system. One of the postulates of quantum mechanics is that everything we can know about a system is summed up in the system's wavefunction, so microstates in quantum mechanics are just individual wavefunctions. In classical mechanics, a microstate is just a listing of the position and velocity of every point-like particle within the system. [SM 3]
So, depending on your tolerance for super-empirical notions, microstates aren't "real" things. I mean, sure, every system has to be in a particular microstate at a particular time, but no one is going to actually observe that. For most systems, there's the practical problem of being unable to store the data — even a mole's worth of material would require much more than 10^23 bits of data — i.e., not happening in this universe, buddy.
So why talk about microstates at all? Because the things that we can observe — pressure, temperature, and so on — are macrostates linked in a one-to-many coorespondence with several microstates. And if we assume that all microstates within a closed system are equally likely (a rather contentious issue called the ergodic hypothesis) we can begin to make statistical claims about which macrostates are more likely than others. Thanks to the Central Limit Theorem and the rather large number of molecules (6x10^23 per mole) in the typical thermodynamic system, these statistical claims tend to have low variance.
In other words, if you can't see them, invent them.
[SM] Franz Schwabl, Statistical Mechanics 2nd ed., Springer-Verlag 2002.