The focus of the paper is the appropriateness of pragmatic elicitation techniques for generating talk to be used in analyses of talk and social structure. In the best pragmatic elicitation techniques (i) data are generated in situations in which researchers can manipulate variables in the testing of hypotheses, and (ii) speakers can talk freely and spontaneously without awareness that their talk is the object of study. This claim was tested in an examination of the hypothesis that more facework will occur in refusals to a High versus Low status requester. Requester status was manipulated in Oral and Written Discourse Completion, Role Play, and an Experimental elicitation technique. Support for the hypothesis was found only in the Role Play and Experimental conditions. Next, refusals generated in the above four elicitation conditions were compared to Naturally-occurring refusals. At the levels of the acts by which refusals are accomplished and the internal structure of the head act, Oral and Written DC produced anomalous and non-representative refusals. Role Play and the Experimental technique produced refusals that were very similar to Natural refusals, though Role Play refusals tended to be somewhat repetitive and long-winded. It is concluded that an Experimental technique is the preferred pragmatic elicitation technique.