Earlier today, I was wiping my daughter Mei's face, and I said she had some "schmutz" (/Smuts/) on it. Being a talkative little three-year-old, she immediately started repeating the word to tell her mom what was going on. I barely caught it, but I noticed that she actually pronounced the word more like "smuch" (/smutS/). This charming little slip of the tongue is called metathesis. (It would be a spoonerism if it happened across at least two words, but I think it's too small to qualify, really.)
You get what happened? It looks like the initial /S/ and the final /s/ traded places, right? Or could it be that a single phonological feature (such as [+palatal] or [-anterior]) simply got misplaced? This sounds like a stretch until you consider my next example.
About a week ago, Mei had a slip of the tongue that really tickled my fancy. She was saying the word "Japan" (/dZ@pan/), a word she knows very well and has never had any trouble pronouncing, and it came out like "Chaban" (/tS@ban/)! She immediately corrected herself, confirming that this really was just a production error. In this case, it seems like the voicing feature of the first syllable's onset moved to the second. Or the [-voice] feature of the second syllable moved to the first. However you look at it, some metathesis occurred, but only at the feature level. She could've just as easily said "Pajan" (/p@dZan/), I suppose, but that's not what happened.
This may not be such a big deal, actually. I've just never heard of this sort of thing occurring, but that doesn't make it unheard of. If anyone has seen any analysis of this sort of thing, please point me to it!
So was the first example (/Smuts/ and /smutS/) a case of segment switching or feature switching? The fact that it's impossible to say based only on the observable evidence should be a clue that this isn't a very productive question. Phonological segments seem to be nothing more than convenient groupings of features anyway. Just as it's hard to say what exactly constitutes a word, the definition of a segment gets a little fuzzy under scrutiny too. The more time passes, the more I think an approach like I outlined in "Active Phonology" is headed in the right direction.