Quite a few things have got me thinking about this recently.
One of the TNIV translators (randomly) got back to my comment on the translation of Galatians 2:17. I still think I'm right, for the simple reason that the NIV, ESV, etc say what the Greek says and the TNIV doesn't.
And Psalm 8, which is a critical passage in Bible translation, has been knocking around in my head for a while.
Type of Translation
The main types of translation are the near-literal translations (ESV, NKJV, NASB, NRSV, Nick King often but not always), the dynamic equivalent translations (NIV, TNIV) and the paraphrase translations (CEV, NLT, Message). Mark Driscoll also lists distortions like the JW one.
As far as I can tell, this should simply be a question of readability versus accuracy. You should go for the version that is as literal as possible to the original while still making sense. For me that means near-literal as often as possible, though in a church context I'm happy with dynamic equivalent and in a culture with low levels of literacy I'd happily go for paraphrase.
The people who are big fans of the dynamic equivalent translations sometimes try arguing that the advances in scholarship mean that we have a greater understanding of what the original writers meant, so that we can put that into better English than a literal word-for-word translation. One problem with that are partly that so often the dynamic equivalent translations lose a lot of the subtleties of the Greek - for example the use of the morphe stem in Philippians, which is key to some of the meaning of the book. The other problem is that what you're actually reading is what the translator thinks God is saying whereas at least with the near-literal translations you have fewer layers of interpretation added onto the text.
One of the big contentious issues in translation is gender neutrality in language. It used to be that "a man" could mean either "a person of indeterminate gender" or "an adult of male gender". But that isn't necessarily true any more. In addition, Greek has three words - aner: "an adult of male gender"; gune: "an adult of female gender" and anthropos: "a person of indeterminate gender" (but it takes masculine pronouns). The old translations all used "man" and "he" for anthropos as well as for aner. Several newer ones (NRSV, TNIV) try to translate anthropos in a gender-neutral way and tend to use "they" for the pronoun. That's good; it's gender neutral in modern English, which the original is too.
The problem comes with some of the translations. For example "Son of man" is huios tou anthropou, so some of the times (like in Psalm 8) when it doesn't clearly refer to Jesus, it's translated in a gender neutral way - "mortal humans" or something. The problem is that Hebrews then picks up Psalm 8 and does apply it to Jesus.
Another problem is the problem of pluralisation. The gender-neutral translations tend to pluralise because it's easier to be gender neutral then. That's a problem in Psalm 1 for example, where there's a clear distinction between the righteous person and the wicked people. But you can't see that at all in the NRSV / TNIV. Gender neutrality where the original is, good. Mangling the Bible, bad.
I think, for example, that the phrase "son of man" needs to be translated consistently. The problem is that to do that in a gender neutral way would mean changing one of the best-known titles of Jesus.
I think it's important to be clear that any of the following translations (and some others) are great and will give you a very good idea of what the original is saying at least 95% of the time. ESV, NASB, NKJV, NIV, TNIV, NRSV. If you're fussy about meaning, it's a good idea to have two from different backgrounds to compare - I'd probably go for ESV and TNIV and say that between them you've pretty much always got a good translation.