I confirm the presence of a serious memory leak in ElementTree, using the iterparse() function.
Memory grows disproportionately to dozens of GB when parsing a large XML file.

For further information, see discussion in:
  http://www.gossamer-threads.com/lists/python/bugs/912164?do=post_view_threaded#912164
but notice that the comments attributing the problem to the OS are quite off the mark.

To replicate the problem, try this on a Wikipedia dump:

    iterparse = ElementTree.iterparse(file)
    id = None
    for event, elem in iterparse:
        if elem.tag.endswith("title"):
            title = elem.text
        elif elem.tag.endswith("id") and not id:
            id = elem.text
        elif elem.tag.endswith("text"):
           print id, title, elem.text[:20]

Can you specify how you import ET? I.e. from the pure Python or the C accelerator?

Also, do you realize that the element iterparse returns should be discarded with 'clear'? [see tutorial here: http://eli.thegreenplace.net/2012/03/15/processing-xml-in-python-with-elementtree/]

Can this be reproduced in 3.2/3.3?

You are right, I should discard the elements.

Thank you.