MOND achieves impressive fits to most galaxies rotation curves with a single adjustable parameter - the mass to light ratio for the star population. There is some correlation between the visible and the required dark matter in spiral galaxies, the most well known example is the Tully-Fisher relation. Tully-Fisher is explainable in MOND but remain mysterious within the dark matter theory. Of course nothing of the above proves MOND is correct. It can be regarded as some empirical formula that produces good fits.
The Bekenstein's relativistic variant of MOND is rather contrived and singles out a preferred frame of reference. MOND fits well the galaxy rotation curves but not the galaxy cluster masses.
If dark matter is not detected in the next 20 years we seriously have to start thinking of alternative theories.