Unfortunately, his "Introduction to Econometrics," now in its third edition, is, at best, a mediocre principles textbook. While his explanations of crucial ideas, such as least squares methodology, heteroskedasticity, or autocorrelation, are concise and extremely clear, a student who wants a more rigorous mathematical introduction to econometrics will be searching for another book. Linear algebra, which makes a number of proofs and tests far simpler and less burdensome to follow than calculus, is resigned to pages in the dark recesses of the appendix. Although these incidentals are well written, they remain peripheral, when they should be the focus of any undergraduate econometrics textbook.
To its credit, Maddala's text does an excellent job of explaining concepts and problems in plain English, which would do well to supplement a purely mathematical approach. The book is full of techniques and tricks that might be helpful to reduce problems of serial correlation and heteroskedasticity, and the later chapters do a very good job of introducing students to more advanced topics, such as panel data analysis and vector autoregression. Nevertheless, the book's treatment of time-series analysis is scattered and atrocious, and many students will find themselves searching for other texts especially for information on this area.
For students who do not mind a bit of mathematical simplicity and would rather seek to grasp the general ideas behind linear regression and its difficulties, G.S. Maddala's "Introduction to Econometrics" should be a good read. Yet, for students who want to pursue their knowledge of econometrics further, this book will most likely not be very helpful. As many courses taught at an undergraduate level tend to vary on their level of rigor, it probably will not hurt to have this book, but it may not help.
I would have no reservation recommending this book to other readers. It is certainly better that Gujarati's 'Basic Econometrics.' However, I think Wooldridge's 'Introductory Econometrics' is a better choice for the beginner, especially in it's coverage of cross section and panel data and its abundance of examples. Unfortunately, since Maddala passed away before the third edition was completed the last three chapters, including the chapter on panel data, do not contain any exercises.
To sum up: An excellent text overall, with some minor shortcomings.
In sum, I think the book is sometimes too elementary to be used for advanced students, while it was too advanced to be used alone as an introductory textbook. I would recommend using/reading selected chapters (such as ch. 12 on model selection), which are accessible as well as dealing with topics not normally included in many textbooks.