TREC, the Text Retrieval Conference, is the information retrieval (IR) community’s annual evaluation forum, sponsored by the US Department of Defense and the National Institute of Standards and Technology (NIST). The event is split into tracks (eg, ad hoc retrieval, filtering, question answering) that encapsulate different research agendas in the community. The end result of each track meeting is an overview report written by the track organizers and a collection of technical reports by the track participants. Many of these reports, after some refinement, find their way into leading IR-related conferences such as SIGIR, and every few years a special issue dedicated to a particular TREC or a TREC track is published. The purpose of the present book is fourfold: to collate and distill 12 years’ worth of experiments (1991–2003) into a single volume; to provide some historical perspective on the evolution of the tasks; to share some of the general findings across tracks; and to encourage participants to take an introspective look at their progress and ask the question, What next for TREC? Despite TREC’s obvious focus on ad hoc retrieval (ie, given a query return a ranked list of relevant documents), this book has a surprising amount to offer the natural language processing (NLP) community, particularly to researchers interested in question answering (QA) and text summarization, and to a lesser extent researchers concerned with the application of information extraction (IE), machine translation, speech processing, and language-generation technologies. It must be stressed, however, that this is not a book for readers looking for an introduction to IR concepts; there are many adequate textbooks that already fill this need such as that of Baeza-Yates and Ribeiro-Neto (1999). Instead it should be viewed as a starting point for researchers who are using standard IR techniques, such as passage retrieval or a term-weighting function, and would like to investigate the state of the art as determined by TREC’s evaluation results. In this review, I will make reference to these NLP interests where appropriate. The book consists of three parts: The first provides a three-chapter overview of TREC, structured around its different tracks, its evaluation methodology, and its test collections; the second consists of seven chapters on selected track reports; and the final part contains seven reports from the perspective of the participants, many of whom have devoted their efforts to multiple TREC tasks over the years. All of these contributions are of a high quality; this is not surprising given that most of the participants have been working on their respective areas for at least the duration of their TREC track (s). Each chapter is followed by its own bibliography, and a comprehensive 12-page index at the back of the book contains entries for keywords and referenced authors. I came across a few editorial oversights, but nothing that significantly downgrades the quality of this publication.