Data mining techniques have been used in past researches to predict childhood obesity, but the results are still inadequate. The purposes of this paper are to use significant parameters for childhood obesity prediction, to study suitable data mining techniques for childhood obesity predictions, and to propose a hybrid data mining technique. The proposed technique is a hybrid of Naïve Bayesian and decision tree (NBTree) that aimed to increase the accuracy of childhood obesity prediction. NBTree has managed to increase the sensitivity of Naïve Bayesian predictions from 63% to 83% but reduced it specificity from 58% to 53%. In the medical predictions, the sensitivity is often more important than the specificity. As a conclusion, the proposed hybrid technique has increased the accuracy of childhood obesity prediction.