I agree with some of what ialsoagree says, but some of it is wrong. MOND is not contradicted by many observed facts. There are indeed claims to this effect, which I have reviewed extensively - most recently in a review invited by the journal Living Reviews in Relativity: http://relativity.livingreviews.org/Articles/lrr-2012-10/fulltext.html. Very few of these claims hold water, and I am disturbed by how readily some of the more absurd claims are accepted uncritically. The sociology seems to boil down to: if it supports dark matter, it must be right; if it supports MOND, it must be wrong.

I do not understand why ialsoagree asserts that there would be no cosmic microwave background (CMB) in MOND. This is not the case. I correctly predicted the amplitude ratio of the first to second peak in the power spectrum of the CMB using a model motivated by MOND. Nobody else got that right ahead of time. The same model fails to get the third peak right (this debate would be over if it had), but by some strange coincidence the first-to-second peak ratio is still measured to be exactly what I predicted ahead of time.

So yes, MOND has a ways to go, but it has already come farther than seems to be appreciated. ADS: http://adsabs.harvard.edu/abstract_service.html - use it to search the literature.

I do agree that MOND suffers problems in clusters, and have said so many times. It is not the only piece of evidence. Some evidence favors LCDM (our current standard cosmological model), some favors MOND. What conclusion you come to depends on how you weigh the evidence. See http://astroweb.case.edu/ssm/mond/LCDMmondtesttable.html. You don't have to agree with my evaluation of the evidence, but I have considered all the things people seem to be fond of asserting automatically exclude MOND. It just isn't that clear.

LCDM has enormous trouble doing the things MOND does easily. Where MOND explains galaxy dynamics with a single free parameter per galaxy, the unavoidable conversion from measured starlight to the mass of the stars producing it, LCDM requires a minimum of three parameters. That doesn't really work, so we invoke "feedback" which brings at least another three free parameters Even then it still doesn't manage as well as MOND. I'm sure we can keep adding free parameters until we get it right, but this is a pretty blatant violation of Occam's razor. If I were to adopt the same attitude as MOND's critics, I would assert that it is impossible for LCDM to do what the data require and therefore it is falsified. That is no less fair a statement than the legitimate complaints about MOND. So again, I don't think it is that easy to choose.

And yes, MOND is a scientific theory, in the same sense that Newton's Law of Universal Gravitation is. MOND is an extension of that theory. In the modern parlance, it can be derived from a Lagrangian. What MOND is NOT is an extension of General Relativity. Reconciling the two is a big challenge for theorists. So far we haven't been able to reconcile General Relativity with the other fundamental forces in a would-be theory of everything, or even quantize gravity satisfactorily. So I hesitate to say there can be no surprises in gravitational theory.

Dark Matter is a concept, not a theory. We can develop specific hypotheses for what the dark matter can be, and construct legitimate theories therefrom. But "dark matter" by itself could be just about anything, and is not subject to falsification. Didn't see enough brown dwarfs? Must be WIMPs. No WIMPs? OK, Axions then. If not that, we can always make up something else.

This subject always brings out the big questions. All we said in the paper (and the press release) was that we applied MOND to the newly discovered dwarfs of Andromeda, and it seems to work. Didn't have to. We also make some a priori predictions that have yet to be tested, so maybe it will still fail. It that not interesting?