Always remember that with the default parameters this function doesn't handle well large files, i.e. if a text node is longer than 10Mb it can raise an exception stating:

DOMDocument::loadXML(): internal error Extra content at the end of the document in Entity

even though the XML is fine.

The cause is a definition in parserInternals.h of lixml:
#define XML_MAX_TEXT_LENGTH 10000000

To allow the function to process larger files, pass the LIBXML_PARSEHUGE as an option and it will work just fine:

$domDocument->loadXML($xml, LIBXML_PARSEHUGE);

Possible values for the options parameter can be found here:

http://us3.php.net/manual/en/ref.libxml.php#libxml.constants

loadXml reports an error instead of throwing an exception when the xml is not well formed. This is annoying if you are trying to to loadXml() in a try...catch statement. Apparently its a feature, not a bug, because this conforms to a spefication. 

If you want to catch an exception instead of generating a report, you could do something like

<?php
function HandleXmlError($errno, $errstr, $errfile, $errline)
{
    if ($errno==E_WARNING && (substr_count($errstr,"DOMDocument::loadXML()")>0))
    {
        throw new DOMException($errstr);
    }
    else 
        return false;
}

function XmlLoader($strXml)
{
    set_error_handler('HandleXmlError');
    $dom = new DOMDocument();
    $dom->loadXml($strXml);    
    restore_error_handler();
    return $dom;
 }

?>

Returning false in function HandleXmlError() causes a fallback to the default error handler.

A call to loadXML() will overwrite the XML declaration previously created in the constructor of DOMDocument. This can cause encoding problems if there is no XML declaration in the loaded XML and you don't have control over the source (e.g. if the XML is coming from a webservice). To fix this, set encoding AFTER loading the XML using the 'encoding' class property of DOMDocument. Example:

Bad situation:

test.xml:
<test>
    <hello>hi</hello>
    <field>ø</field>
</test>

test.php:
$xmlDoc = new DOMDocument("1.0", "utf-8"); // Parameters here are overwritten anyway when using loadXML(), and are not really relevant
$testXML = file_get_contents("test.xml");
$xmlDoc->loadXML($testXML);
// Print the contents to a file or in a log function to get the output, using $xmlDoc->saveXML()

Output:
<?xml version="1.0"?>
<test>
    <hello>hi</hello>
    <field>&#xF8;</field>
</test>

Good situation:

test.xml:
<test>
    <hello>hi</hello>
    <field>ø</field>
</test>

test.php:
$xmlDoc = new DOMDocument("1.0", "utf-8"); // Parameters here are overwritten anyway when using loadXML(), and are not really relevant
$testXML = file_get_contents("test.xml");
$xmlDoc->loadXML($testXML);
$xmlDoc->encoding = "utf-8";
// Print the contents to a file or in a log function to get the output, using $xmlDoc->saveXML()

Output:
<?xml version="1.0" encoding="utf-8"?>
<test>
    <hello>hi</hello>
    <field>ø</field>
</test>

earth at anonymous dot com,

preserveWhiteSpace property needs to be set to false for formatOutput to work properly, for some reason.

$dom = new DOMDocument;
$dom->preserveWhiteSpace = false;
$dom->loadXML($xmlStr);
...
$element->appendChild(...);
...
$dom->formatOutput = true;
$xmlStr = $dom->saveXML();
echo $xmlStr;

This would format the output nicely.

The documentation states that loadXML can be called statically, but this is misleading. This feature seems to be a special case hack and its use seems to be discouraged according to http://bugs.php.net/bug.php?id=41398.

Calling the method statically will fail with an error if the code runs with E_STRICT error reporting enabled.

The documentation should be changed to make it clear that a static call is against recommended practice and won't work with E_STRICT.

Note that loadXML crops off beginning and trailing whitespace and linebreaks.

When using loadXML and appendChild to add a chunk of XML to an existing document, you may want to force a linebreak between the end of the XML chunk and the next line (usually a close tag) in the output file:

$childDocument = new DOMDocument;
$childDocument>preserveWhiteSpace = true;
$childDocument->loadXML(..XML-Chunk..);
$mNewNode = $mainDOcument->importNode($childDocument->documentElement, true);
$ParentNode->appendChild($mNewNode);
$ParentNode->appendChild($mainDocument->createTextNode("\\n  ");

Although it is said that DOM should not be used to make 'pretty' XML output, it is something I struggled with to get something that was readable for testing.  Another solution is to use the createDocumentFragment()->appendXML(..XML-Chunk..) instead, which seems not to trim off linebreaks like DOMDocument->loadXML() does.

For some reason, when you set DOMDocument's property 'recover' to true, using '@' to mask errors thrown by loadXml() won't work.

Here's my workaround:

function maskErrors() {}
set_error_handler('maskErrors');
$dom->loadXml($xml);
restore_error_handler();

You could also simply do this: error_reporting(0); and then set back error_reporting to its original state.

When using loadXML() to parse a string that contains entity references (e.g.,  ), be sure that those entity references are properly declared through the use of a DOCTYPE declaration; otherwise, loadXML() will not be able to interpret the string.

Example:
<?php
$str = <<<XML
<?xml version="1.0" encoding="iso-8859-1"?>
<div>This&nbsp;is a non-breaking space.</div>
XML;

$dd1 = new DOMDocument();
$dd1->loadXML($str);

echo $dd1->saveXML();
?>

Given the above code, PHP will issue a Warning about the entity 'nbsp' not being properly declared.  Also, the call to saveXML() will return nothing but a trimmed-down version of the original processing instruction...everything else is gone, and all because of the undeclared entity.

Instead, explicitly declare the entity first:
<?php
$str = <<<XML
<?xml version="1.0" encoding="iso-8859-1"?>
<!DOCTYPE root [
<!ENTITY nbsp "&#160;">
]>
<div>This&nbsp;is a non-breaking space.</div>
XML;

$dd2 = new DOMDocument();
$dd2->loadXML($str);

echo $dd2->saveXML();
?>

Since the 'nbsp' entity is defined in the DOCTYPE, PHP no longer issues that Warning; the string is now well-formed, and loadXML() understands it perfectly.

You can also use references to external DTDs in the same way (e.g., <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
"http://www.w3.org/TR/html4/strict.dtd">), which is particularly important if you need to do this for many different documents with many different possible entities.

Also, as a sidenote...entity references created by createEntityReference() do not need this kind of explicit declaration.

Instead of doing this:

<?php
$str = <<<XML
<?xml version="1.0" encoding="iso-8859-1"?>
<!DOCTYPE root [
<!ENTITY nbsp "&#160;">
]>
<div>This&nbsp;is a non-breaking space.</div>
XML;

$dd2 = new DOMDocument();
$dd2->loadXML($str);

echo $dd2->saveXML();
?>

simply use:

loadHTML() rather than loadXML().

While loadXML() expects its input to have a leading XML processing instruction to deduce the encoding used, there's no such concept in (non-XML) HTML documents. Thus, the libxml library underlying the DOM functions peeks at the <META> tags to figure out the encoding used.

See http://xmlsoft.org/encoding.html.