I'm going to have to disagree with a lot of people here (including myself, 5 years ago). Some time ago, I did a series of controlled tests to see what search engines indexed and what they did not. I made up a fake word, added some variations of it to test stemming and internationalization, and then paid close attention to the results.
The reason I did the tests was to check international spelling issues - for example, how does Google treat pages talking about "Montréal" (with accent)when a searcher types in "Montreal" (without accent). Since my specialty is international SEO, this is a big deal for me and I assure you I was very careful to make sure that the tests were useful and applicable to my work. It wasn't just for fun.
I stuffed those fake words pretty much anywhere you could think of, and even a few places the spammers have usually missed, just to test what Google and friends index and what they don't. It was a very interesting set of experiments and highlighted some important differences between search engines.
The test also clearly showed that keywords in the URL are indexed and count towards ranking. I was not able to determine how much they count, but it's more than pure inclusion (such as Yahoo's handling of the keyword metatag).
This is not to say that the anchor text factor isn't important - I believe it's much more important than the URL. But I have no doubt that having keywords in your URL help search engines understand the intended topic of your page better, and thus result in a stronger tendency to rank, everything else being equal.
And if you think about it, why wouldn't it? Is there some reason why a search engine would decide that a page called "dog-houses.htm" was irrelevant to a searcher looking for dog houses? Any more irrelevant than an ALT attribute, H1, noscript or image name?
No, search engines look at a lot of things in order to figure out content, clarity and context, and it doesn't surprise me in the slightest that they would look at the name of a file (which is really what a URL is) as one of the things they look at to determine the topic of that same file. And my tests support that. If someone has a test that they can show me that shows that it doesn't matter (not just anecdotal evidence or opinion) I'd love to see it, since if my tests or conclusions are wrong I'd like to know how and why. But I don't think they are, in this case.
Edited by mcanerin, 14 November 2008 - 12:15 PM.