As natural language processing systems grow larger, they need larger lexicons. Furthermore, larger vocabularies require larger, more complex lexical entries to distinguish between different senses and uses. The machine-readable dictionary is a large, compact source of lexical information, but conversion of its contents to a form that a computer can use is an immense task. This thesis describes an effort to extract semantic information from Webster's Seventh New Collegiate Dictionary (W7) by a variety of methods, including parsing with Sager's Linguistic String Parser and semi-interactive text processing using UNIX utilities. It compares the effectiveness of those methods for different kinds of analysis and suggests analytical strategies exploiting the strengths of each method. Also discussed are the nature of the language of W7 and some issues in the design of a lexical database.