As the vehicle of data collection, the questionnaire is one of the critical components in achieving high quality in a survey. The best of sampling schemes and estimation strategies will not yield accurate data if the answers provided by the respondent are not meaningful. In this paper, we describe three of the methods that are used to pretest questionnaires: cognitive interviewing, respondent debriefing, and behavior coding of respondent/interviewer interaction. We recognize that there are several other valuable pretesting methods such as expert panels, questionnaire appraisal coding systems, interviewer debriefings, etc. The methods we discuss were chosen since together they provide information from all three potential sources of measurement error: the questionnaire, interviewer, and respondent. While these three methods are used widely throughout the Federal government and by other data collectors as well, we will concentrate on their use at the Census Bureau. We present brief descriptions and historical summaries of each of these methods, as well as examples of their use, the results they obtain, and how they can be used to improve questionnaires.