Purpose - In survey interviews information is transferred to the researchers via a communication process between interviewers and respondents. This process is controlled directly by the interviewers, and indirectly by the researchers who constructed the questionnaire and instructed and supervised the interviewers. In spite of these control activities, errors occur. This paper investigates the sources of these errors. Design/methodology/approach - In order to investigate the sources of these errors, transcripts of 200 interviews were analyzed using a detailed coding scheme. Findings - In 30 percent of all question-answer sequences interviewer and respondent stick to the "script" designed by the researcher. In these "paradigmatic" sequences the open loop control by the researcher works well. In 25 percent of the sequences this control is not sufficient, but additional closed loop control, via "repair" activities of the interviewers, appears to be successful. In the remaining sequences both the open loop control of the researcher and the closed loop control by the interviewer failed. Originality/value - The recently developed systematic analysis of question-answer sequences in survey interviews, employed in this research, offers detailed insight into the errors occurring during the interview process, and illustrates the need for improved question design and improved training of interviewers.