The aim of this study was to assess the reliability and validity of three different observation-based measures of patient-centredness. The three face-valid instruments were each applied to the same sample of 55 videotaped GP consultations. Associations were explored with consultation 'input' variables (e.g. patient and doctor demographic characteristics, patient health status) and 'process' variables (e.g. consultation length). The three measures demonstrated varying levels of inter-rater reliability. Reliability was proportional to training requirements. Differences in construct validity of the three measures were evident and their concurrent validity was relatively low. Researchers must exercise caution in their choice of measurement method because of differences in how the concept of 'patient-centredness' is operationalized. Greater conceptual specificity and simplification are required for meaningful, reliable measurement. The implications for research, and for assessing the quality of individual doctors' 'interpersonal' care are discussed.