Objective To evaluate the reliability of anthropometric measurements (weight, height, Body Mass Index (BMI), waist and hip circumferences (WC; HC) and waist-to-hip ratio (WHR)) performed by doctors to assess obesity. Method Repeated anthropometric measurements were performed by 12 primary care physicians on 24 adult volunteers in Geneva, Switzerland, 2006. Volunteers (54% women, mean age 41) had a mean BMI of 28.1 (respective mean values for WC, HC and WHR: 91.4, 108.3, 0.84). Inter-observer reliability coefficient (R) and percent disagreement in categorisation of volunteers (normal weight, overweight, obesity, abdominal obesity) were computed according to these measurements. Results The inter-observer reliability for weight, height, and derived BMI were excellent (R > 0.99), but unsatisfactory for WC (R = 0.92), HC (R = 0.76) and WHR (R = 0.51). Based on the BMI, only 1% of the volunteers were misclassified as overweight or obese, whereas the use of WC and WHR lead to misclassification in 6% and 23% respectively. Reliability for the measurements improved after a one-hour training in anthropometric measurements (R = 0.97 for WC, 0.92 for HC and 0.89 for WHR), but the proportion who were misclassified remained high despite the training session for WC (5%) and WHR (9%). Conclusions BMI remains the most reliable measure to detect obesity in medical practice, whereas WC, HC and WHR are less reliable. These results challenge current recommendations on obesity-related cardio-vascular risk management based on WC and WHR and underline the need for further research to improve the reliability of anthropometric measurements by doctors.