iZotope RX is absolutely wonderful for this sort of thing. The denoise section of it works well on its own, and you can do spectral edits (like in SPEAR), but with much better detail, accuracy, and results. I think there's a trial available on the website.
EQ is also your friend here, but it's sometimes difficult to isolate stuff without losing necessary frequencies. If you're dealing with a siren, you can probably roll off a lot of the low frequencies and not lose anything you need (just a guess, albeit).
I've had great results with it in my often less-than-ideal recording environments.
That said, it really depends on how pure you need the sound to be. Any sound isolation is going to lose something in the edit, especially if it's in a city. It also depends on how much the siren stands out from the background, how loud it is, and what's going on in the background. I wouldn't expect perfect results with city isolation in general, though.
This thread is also going on at the moment:
http://socialsounddesign.com/questions/69/sound-off-on-noise-reduction. It might be of some help.