OBJECTIVE: The extent of intra- and inter-interviewer variability both in anthropometric measurements and in estimates of body composition was assessed and the possibility of systematic variation due to interviewer differences investigated. DESIGN AND SUBJECTS: Seventeen interviewers trained in the anthropometric measurement technique and 10 healthy volunteers (4 men, 6 women) participated in the study on measurement variability. To ensure participation of all interviewers the study was carried out on three different days. On each of these days interviewers got randomly allocated to the subjects being present. Each interviewer took 12 measurements--body weight, body height, sitting height, circumferences of waist, hip, and midarm, skinfolds (biceps, triceps, subscapular, and suprailiac), chest breath and depth--per subject on two occasions. From these measurements, body mass index, waist-to-hip ratio, percentage of body fat, fat mass, fat free mass and metric index were determined. For all anthropometric variables variance components, reliability coefficients (R) and coefficients of variation (CV) were estimated and systematic differences of measurements between interviewers were assessed. RESULTS: Measurement reliability in basic anthropometric measures expressed as variance components, reliability coefficients and coefficients of variation was influenced to a greater extent by inter-interviewer variability (R: 0.858-0.999; CV: 0.1-20.9) than intra-interviewer variability (R: 0.979-0.999; CV: 0.0-6.4). The respective estimates of body composition exhibited comparatively higher reliability (Rinter: 0.975-0.999; Rintra: 0.995-0.999). Measurements more prone to subjectivity, e.g. skinfolds showed lower reliability (CVinter: 9.3-20.9; CVintra: 3.6-6.4). Although the absolute variation in measures due to interviewers was small, systematic differences among interviewers were clearly evident in all measurements and estimates except sitting height in this group of subjects. CONCLUSION: Anthropometric measures and estimates of body composition obtained in the current study show the feasibility of detailed anthropometric data collection by multiple interviewers in large-scale epidemiological studies.