The assessment of weight status is important in many epidemiological studies, but its direct measurement is not always possible. Self-reported weight and height are often used, although previous research reported low accuracy. This study aimed to test the ability of trained observers to accurately estimate weight status in adults using structured observation. A cross-sectional study was conducted. For each participant, height and weight were estimated in categories, and weight status was recorded using Stunkard’s body figures, by two trained observers. Height and weight were also measured, using standardized procedures. Subjects were classified according to World Health Organization body mass index (BMI) cut-offs from objective measurements and from the BMI assigned to each body figure. Sensitivity, specificity, and likelihood ratios were calculated to assess the accuracy of estimating weight status by observation. Kappa was used to test inter-observer reliability. A total of 127 participants were assessed, 70 women and 57 men, aged between 19 and 89 years (mean ± standard deviation: 50.3 ± 16.3 years). Most participants were overweight or obese (64.3% women; 78.9% men). The sensitivity and specificity of overweight/obesity status identification were 72.8% and 78.4%, respectively. Observers’ gender, participants’ gender, and participants’ age were significantly associated with the estimation of overweight/obesity. The agreement between observers was moderate for BMI estimates (κ = 0.52) but substantial when distinguishing normal weight from overweight/obesity (κ = 0.67). Trained observers were able to distinguish normal weight from overweight/obesity with high sensitivity and specificity, and substantial interrater reliability. This innovative methodology showed potential for improvement through enhanced training techniques. The use of structured observation may be a useful and accurate alternative to self-reported weight status assessment, whenever anthropometric measurement is not achievable.