A special-purpose corpus serves, within terminology tasks, as a basis for performing several functions, which can be reduced, basically, to four: a) to identify candidate terms; b) to provide more data about these possible terms (combinations, derivatives, relations with other terms, etc.); c) to help compile and elaborate definitions; d) to offer an important number of contextual examples. However, the first problem a terminologist may encounter when retrieving data from a raw corpus is how to proceed methodically so as to be able to isolate terms relevant to the subject-matter field. The objective of this article is, firstly, to present a four-step methodological proposal to analyze such a corpus; secondly, to provide a systematic way to use certain computer instruments, which are also described; and, thirdly, in combination with these instruments, to show a series of strategies for the semiautomatic recognition and extraction of terminology.