If you’re on the iPhone, using tree-based parsing can be a prohibitive memory hog. Trust me, I’ve been there, and I’ve tried many different approaches over the last five months of development of my main iPhone application. Tree-based parsing works fine until you download someone’s comment stream containing 400 very long comments, clocking in at about 600KB of raw data. Quite aside from the size of the resultant XML tree, the memory allocated internally while creating that tree can be enormous.
I wound up creating a variant of NSXMLParser which pulls data in from a supplied NSInputStream rather than using a single chunk of data, and which passes only 1KB at a time into libxml for handling (NSXMLParser uses libxml too, but passes 100% of the data in one go).
The source code is available on github (look in the StreamingXMLParser folder). You’ll also find a delegate superclass in there; for most parsing needs you can subclass AQXMLParserDelegate and implement -start[Element]WithAttributes: (NSDictionary *) attrs and -end[Element] in your subclass. These methods will be called for you as start & end tags are discovered, and inside the end tag you can use self.characters to access the content characters or CDATA of the element.
For more on the relative memory footprints of the different parsers (albeit on the Mac, not the iPhone) see my original blog post here and the followup on NSXMLDocument here.