Objective To develop a prediction equation for estimation of body fat percentage in elderly populations based on anthropometric data and demographic data, and to verify its prediction power. Methods A total of 1,025 elderly adults were selected by stratified convenience sampling and their anthropometric data （e.g. height, weight and waist circumference, etc.）, demographic data （e.g. gender and age, etc.） and body fat percentage were measured.They were randomized into a model set of 820 and a validation set of 205.Using body fat percentage by bioelectrical impedance analysis （BIA） as the gold standard, stepwise multiple li-near regression was used to develop the prediction equation.ROC curve was used to evaluate the prediction power.The Bland-Altman diagram and concordance correlation coefficient （CCC） were used to verify the consistence of prediction equation and BIA. Results A new prediction equation was developed, based on BMI, gender, height, waist circumference and age.The area under ROC curve was 0.910, the sensitivity was 92.9%, and the specificity was 89.1%.The mean difference between the prediction equation and BIA was -0.021%, and the 95% limits of agreement （LOA） were -6.605% and 6.562% (with a 95% confidence interval of -7.397% to 7.354%).The CCC value between the prediction equation and BIA was 0.883. Conclusion The formula is valid for estimation of body fat percentage in older people and can be used in prevention and management of obesity.