Question answering systems that can understand natural language questions have become a hot item in the world of web search. Despite their increased visibility, there are yet many types of requests that they cannot handle, among which are questions beginning with the word how that require procedural answers. This thesis is meant to be the first of many investigations into the challenges that must be overcome before providing automatic answers to procedural how questions can become a reality. Specifically, we explore the task of distinguishing between procedural and non-procedural web documents. We gathered a corpus, manually classified each document as procedural or non-procedural, devised a set of textual and structural attributes for which statistics were collected, and then applied classification methods to evaluate the degree to which these attributes determined whether the document was procedural. The results of our experiments indicated that the features chosen for classification show promise.