Objective: To measure BMR in a probability sample of adults from an urban city of Brazil and to compare indirectly measured BMR (BMR i ) with BMR predicted from different equations. Design: BMR data were obtained by indirect calorimetry and estimated by different predictive equations (Schofield; Harris and Benedict; Henry and Rees). Anthropometric and body composition measures were also obtained. Setting: The Nutrition, Physical Activity and Health Survey (PNAFS), a household survey conducted in Niterói, Rio de Janeiro state, Brazil. Subjects: Representative sample of 529 adults (aged $20 years; 339 females) living in Niterói, Rio de Janeiro state, Brazil. Results: Mean BMR i values were 5839?7 (SE 73?9) kJ/d and 4758?1 (SE 39?5) kJ/d for men and women, respectively. Predicted BMR by all equations was significantly higher (difference between means and 95 % CI did not include zero) than BMR i in both men and women of all ages. Overall bias in BMR (predicted BMR minus BMR i ) using the Schofield equations (overestimation of about 20 %) was higher than when using the Henry and Rees equations (13 % and 16 % overestimation for males and females, respectively). The percentage of individuals whose BMR predicted by the Schofield equations fell within 10 % of BMR i was very low (7?8 % and 14?1 % of males and females, respectively). Conclusions: Current available predictive equations of BMR are not adequate to estimate BMR in Brazilians living in Niterói, Rio de Janeiro, Brazil.