The aim of this study is to show how to analyze the quality of questions for proxy informants by means of behavior coding. Proxy questions can undermine survey data quality because of the fact that proxies respond to questions on behalf of other people. Behavior coding can improve questions by analyzing interviewer-respondent interactions. Twenty-nine proxies participated in the pretesting of a disability questionnaire. The questionnaire includes 11 questions related to daily-life limitations as a result of health problems. Interviewer-proxy interactions were coded and analyzed by means of Sequence Viewer program. The percentages, from a methodological perspective, of ideal "question-and-answer" sequences varied from 28% to 76% throughout the 11 questions analyzed. The results obtained pointed out the necessity of reviewing some of the proxy questions analyzed. Behavior coding can improve the quality of proxy questions in health surveys when proxy informants are surveyed.