jQuery.expr[':'].data = function(elem, index, m) {
 
    m[0] = m[0].replace(/:data\(|\)$/g, '');
 
    var regex = new RegExp('([\'"]?)((?:\\\\\\1|.)+?)\\1(,|$)', 'g'),
        key = regex.exec( m[0] )[2],
        val = regex.exec( m[0] )[2];
 
    return val ? jQuery(elem).data(key) == val : !!jQuery(elem).data(key);
 
};

Its usage is as follows:

$('div').data('foo','bar');
 
// Select all elements with "foo" set to "bar":
$('*:data(foo,bar)')
 
// With quotes (for complex strings):
$('div:data("foo","b a r")');

@Kyle, I agree; from a code maintainability point of view, the previous data selector is definitely superior. Maybe we could meet halfway with a ‘:’ or even a ‘data:’ prefix like Balazs suggested? That would definitely make it more explicit:

jQuery('a[data:ABC=123]');

Plus, the new implementation ties in very nicely with Sizzle’s core and so is slightly faster than the older ‘:data()’ one. It’s really a matter of taste in the end… go with whatever seems more intuitive.

there’s already a ticket for it as well:
http://dev.jquery.com/ticket/3586

and I also made a plugin a few months ago that does this a bit differently:
http://plugins.jquery.com/project/EnhancedAttributeSelectors

Anyway, this implementation is very elegant, and doesn’t add as much overhead as mine at all. What I would only suggest is using a prefix so there’s no ambiguity if you’re dealing with an attribute or data:

            $.expr.attrHandle[':' + key] = function(elem){
                return $(elem).attr(key) || $(elem).data(key);
            };

then you can select by data like this: $('a[:ABC=123]');

Jack.

If I had to jump into someone elses code and they were using this, I probably would go crazy trying to figure out how they are getting values from attributes that aren’t there. I’d probably have no clue that they have this extra little snippet at the bottom of the js file somewhere.