This book has been a great surprise to me. Initially I thought that in about 300 pages (excluding homeworks and exercises) I could not find all I could need for an Automata, Languages and Computation course. I was wrong, definitely. The book is coincise, but also rich and precise.
The material is very well chosen, and the writing stile is directly thought with students in mind. Kozen has a pluri-annual experience in teaching at Cornell University, and it seems he has developed an effective style of communication with students, that's perfectly reflected in his books.
Some important topics are present in this book and not in both Sipser and Hopcroft-Ullman. If you need (as I did) to learn about Myhill-Nerode Relations and Theorem, this book features the best account I've seen (the other, much shorter, reference can be found in the first editon of Hopcroft-Ullman but not in the second one !).
A nice shot of the Lambda-calculus is also featured, and this too lacks in the other two books.
The organization in lectures is a very good idea when studying. Lectures are carefully cut and self-contained, so that you can organize your time using this unit, and wherever you choose to stop a study session, you always stop at correct boundary of a topics.
As a further (and important) note, the notation used is very clear and elegant. As soon as you get used with it (very soon since its clarity) it becomes very stimulating. Don't understimate this value, since many books feature too-hard-to-follow notations, or no notation at all. Both of which cases are to be avoided, INMH.
I have used other books for my course, starting from both the editions of the Hopcroft and Ullman, but one way or the other I found myself always with this book (and Sipser's) in my hands.