Empirical Methods for Artificial Intelligence by Paul Cohen
ISBN 0-26-20322-52 A wonderful book which, while it has a slightly different notion of science (from my own), nonetheless talks a lot about how we can be more scientific in our work.
Computer science and artificial intelligence in particular have no curriculum in research methods, as other sciences do. This book presents empirical methods for studying complex computer programs: exploratory tools to help find patterns in data, experiment designs and hypothesis-testing tools to help data speak convincingly, and modeling tools to help explain data. -- Amazon review