To efficiently discriminate the promoter in human genome, an algorithm for human promoter recognition based on single nucleotide statistics and Support Vector Machine (SVM) ensemble was proposed. Firstly, a gene dataset was divided into two subsets such as C-preferred and G-perferred subsets by using single nucleotide statistics. Secondly, DNA rigidity feature, word-based feature and CpG-island feature were extracted for each subset. Finally, these features were combined by using SVM ensemble learning. In addition, three ensemble ways were discussed, including single SVM ensemble, double-layer SVM ensemble and cascaded SVM ensemble. The experimental result shows that the proposed method can improve the sensitivity and specificity of human propoter recognition. Especially, the double-layer SVM ensemble can achieve the highest sensitivity of 79.51%, while the cascaded SVM ensemble has the highest specificity of 84.58%.