When testing an antibody for an imaging application, it is almost always a good idea to test it in another application like ELISA or Western blotting to see if it binds the target of interest. For example, try to find high- and low-expressing cell lines, load equal amounts, and see if there is a difference in signal at the expected molecular weight. Check the company's literature or contact their technical support to see if they have tested the antibody by immunofluorescence (IF), and if so what their results were and how they validated them. Check the literature to see if other people have published with this product. http://highwire.stanford.edu is a pretty good full-text search engine, but they don't have all journals in their database. Google Scholar can help, too.
You can also use your high- and low-expressing cell lines to see if there is a stronger IF signal in the high-expressing line. Look at your images to see if the signal is localizing where you expect it to, and do co-staining with antibodies to other proteins that should co-localize with your target of interest. For example, if your target is expected to localize to mitochondria, co-stain with COX-IV. A quick Google search found this Cellular Localization IF Antibody Sampler Kit that has controls for several different organelles and other specific locations.
Make sure you are following the manufacturer's recommended protocol. Don't crank the gain up too high on the scope, because non-specific background goes up, too. Look at the induction of signal between high- and low-expressing cell lines. Generally you would want at least a 2-3 fold shift, especially if you see a strong change in signal intensity by Western. Try using siRNA or similar silencing technology to knock down the signal in your cells, and validate using Western and IF.
Ultimately, you can never be absolutely 100% certain that an antibody is binding a certain target, and only that target, but you can accumulate a lot of evidence to show that it is likely. Try to put yourself in the mindset of a reviewer, and objectively convince yourself that the signal is real. If you can't, then shop around and try something else. Good luck!