How can I match on an attribute that contains a certain string?

I am having a problem selecting nodes by attribute when the attributes contains more than one word. For example:

<div class="atag btag" />

This is my xpath expression:

//*[@class='atag']

The expression works with

<div class="atag" />

but not for the previous example. How can I select the <div>?

Here's an example that finds div elements whose className contains atag:

//div[contains(@class, 'atag')]

Here's an example that finds div elements whose className contains atag and btag:

//div[contains(@class, 'atag') and contains(@class ,'btag')]

However, it will also find partial matches like class="catag bobtag".

If you don't want partial matches, see bobince's answer below.

mjv's answer is a good start but will fail if atag is not the first classname listed.

The usual approach is the rather unwieldy:

//*[contains(concat(' ', @class, ' '), ' atag ')]

this works as long as classes are separated by spaces only, and not other forms of whitespace. This is almost always the case. If it might not be, you have to make it more unwieldy still:

//*[contains(concat(' ', normalize-space(@class), ' '), ' atag ')]

(Selecting by classname-like space-separated strings is such a common case it's surprising there isn't a specific XPath function for it, like CSS3's '[class~="atag"]'.)

try this: //*[contains(@class, 'atag')]

EDIT: see bobince's solution which uses contains rather than start-with, along with a trick to ensure the comparison is done at the level of a complete token (lest the 'atag' pattern be found as part of another 'tag').

"atag btag" is an odd value for the class attribute, but never the less, try:

//*[starts-with(@class,"atag")]

A 2.0 XPath that works:

//*[tokenize(@class,'\s+')='atag']

or with a variable:

//*[tokenize(@class,'\s+')=$classname]