Manning[1], the Pragmatic Bookshelf[2], and Apress[3] have been the best in my experience. They cover relevant topics and technologies, the books are typically written by experts and written well, there are ebook versions of everything at reasonable-ish pricing, and sometimes you can get early access to books before their finished.

1: http://manning.com/ 2: http://pragprog.com/titles 3: http://www.apress.com/

I really like books from Manning publications. Their "in action" series are excellent.

If Syngress isn't bad then they've definitely improved. For a while they had some great security titles, but the mid/later half of the last decade they produced a lot of bad books. Terrible editing, repeating chapters used in different books, etc. I don't think Ive bought one of their titles in the last 4-5 year and I haven't regretted it.

It's really alien to me to judge a book based on its publisher. Whenever I need a book I check out reviews on Amazon or on programming sites to find the best in class. Is there a reason that it matters that I'm not thinking of?

It mattered a lot before. If you had no online reviews to go by, going into a book store and picking up an O'Reilly book was a reasonably good bet.

It doesn't matter so much now, and it also only worked because O'Reilly was a small publisher who had the benefit of being able to be really picky about authors and subjects. It doesn't scale when you want to get a large number of books out on subjects where you may have very little choice in authors. How many people are qualified to write a book about [insert random 5 months old javascript library of the week]? Often it might be a handful of people, several of whom won't be interested and/or are bad writers.