As Paul and Mohammed said, the kind of parser you use depends on your data, and what you want to do with it.
If you are "importing" an XML file with many "records" into a database, then you probably want an "event-based parser" like SAX or StAX. You can iterate through the file quickly, transferring the data into database tables. I see these parsers as more low-level, "close to the byte" tools to be used when you really want to crank through a boatload of data.
If you want to do more than this, then more object-based toolkits may be in order, like JAXB, XMLBeans, and Castor. Using these tools you end up with a Java object with the data from the XML. You can then do what you like with the data. One list is here:
http://java-source.net/open-source/xml-parsers
I have used XMLBeans with good success, while JAXB would not compile the schema I needed. And yes, in my case, I have no control over the schema (.XSD and .WSDL files), and even less control over how well my source of XML conforms to it. Having said that, the "lazy object creation" claims I have seen about XMLBeans have not held up for me, and trying to read a large (~800MB) XML document used up all available memory and crashed.
Kind regards,
Drew