The corpus analysis program WordSmith Tools allows two corpora to be compared against each other in order to acquire a list of keywords: words that appear statistically more often in one text than the other. This article examines some potential problems of analyzing keywords and ways to overcome them. A keywords analysis may force the researcher to overplay differences rather than similarities, and it focuses on difference at the lexical rather than the semantic, grammatical, or pragmatic levels. An examination of dispersion patterns, concordances, and key clusters are useful supplementary forms of analysis. The use of annotated data to derive key categories or grouping like words into similar groups is also helpful.