Antimatter is in every precise meaningful sense matter moving backward in time. The notion of "moving backward in time" is nonsensical in a Hamiltonian formulation, because the whole description can only go forward in time. That's the definition of what the Hamiltonian does--- it takes you forward in time a little bit. So if you formulate quantum mechanics the Hamiltonian way, this idea is difficult to understand (still it can be done--- Stueckelberg discovered this connection before the path integral, when field Hamiltonians were the only tool).
But in Feynman's particle path-integral picture, when you parametrize particles by their worldline proper time, and you renounce a global causal picture in favor of particles splitting and joining, the particle trajectories are consistent with relativity, but only if the trajectories include back-in-time trajectories, where coordinate time ticks in the opposite sense to proper time.
Looked at in the Hamiltonian formalism, the coordinate time is the only notion of time. So those paths where the proper time ticks in the reverse direction look like a different type of particle, and these are the antiparticles.
Sometimes there is an idenification, so that a particle is its own antiparticle.
Precise consequence: CPT theorem
The "C" operator changes all particles to antiparticles, the P operator reflects all spatial directions, and the T operator reflects all motions (and does so by doing complex conjugation). It is important to understand that T is an operator on physical states, it does not abstractly flip time, it concretely flips all momenta and angular momenta (a spinning disk is spinning the other way), so that things are going backwards. The parity operator flips all directions, but not angular momenta.
The CPT theorem says that any process involving matter happens exactly the same when done in reverse motion, in a mirror, to antimatter.
The CPT operator is never the identity, aside from the case of a real scalar field. CPT acting on an electron produces a positron state, for example. CPT acting on a photon produces a photon going in the same direction with opposite polarization (if P is chosen to reflect all spatial coordinate axes, this is a bad convention outside of 3+1 dimensions).
This theorem is proved by noting that a CPT operator corresponds to a rotation by 180 degrees in the Euclidean theory, as described on Wikipedia.
Precise consequence: crossing
Any amplitude involving particles A(k_1,k_2,...,k_n) is analytic in the incoming and outgoing momenta, aside from pole and cut singularities caused by producing intermediate states. In tree-level perturbation theory, these amplitudes are analytic except when creating physical particles, where you find poles. So the scattering amplitudes make sense for any complex value of the momenta, since going around poles is not a problem.
In terms of mandelstam variables for 2-2 scattering, s,t,u (s is the CM energy, t is the momentum transfer and u the other momentum transfer, to the other created particle), the amplitude is an analytic function of s and t. The regions where the particles are on the mass shell are given by mandelstam plot, and there are three different regions, corresponding to A+B goes to C+D , Cbar + B goes to Abar+ D, and A + Dbar goes to C+Bbar. These three regimes are described by the exact same function of s,t,u, in three disconnected regions.
In starker terms, if you start with pure particle scattering, and analyticaly continue the amplitudes with particles with incoming momentum k's (with positive energy) to negative k's, you find the amplitude for the antiparticle process. The antiparticle amplitude is uniquely determined by the analytic contination of the particle amplitude for the energy-momentum reversed.
This corresponds to taking the outgoing particle with positive energy and momentum, and flipping the energy and momentum to negative values, so that it goes out the other way with negative energy. If you identify the lines in Feynman diagrams with particle trajectories, this region of the amplitude gives the contribution of paths that go back in time.
So crossing is the other precise statement of "Antimatter is matter going back in time".
The notion of going back in time is acausal, meaning it is excluded automatically in a Hamiltonian formulation. For this reason, it took a long time for this approach to be appreciated and accepted. Stueckelberg proposed this interpretation of antiparticles in the late 1930s, but Feynman's presentation made it stick.
In Feynman diagrams, the future is not determined from the past by stepping forward timestep by timestep, it is determined by tracing particle paths proper-time by proper-time. The diagram formalism therefore is philosophically very different from the Hamiltonian field theory formalism, so much so Feynman was somewhat disappointed that they were equivalent.
They are not as easily equivalent when you go to string theory, because string theory is an S-matrix theory formulated entirely in Feynman language, not in Hamiltonian language. The Hamiltonian formulation of strings requires a special slicing of space time, and even then, it is less clear and elegant than the Feynman formulation, which is just as acausal and strange. The strings backtrack in time just like particles do, since they reproduce point particles at infinite tension.
If you philosophically dislike acausal formalisms, you can say (in field theory) that the Hamiltonian formalism is fundamental, and that you believe in crossing and CPT, and then you don't have to talk about going back in time. Since crossing and CPT are the precise manifestations of the statement that antimatter is matter going back in time, you really aren't saying anything different, except philosophically. But the philosophy motivates crossing and CPT.