Approximately 50% of older patients with acute myeloid leukemia (AML) do not obtain chromosomal abnormalities as an effective risk-stratification, and present cytogenetically normal AML (CN-AML). To develop a reliable prediction model for stratifying the risk of these elderly patients, we conducted a study with a discovery and validation design. As a result, we found the top 6 mutated genes in the discovery cohort of 26 case by the whole exome sequencing, and verified as recurrent mutations in the large cohort of 329 patients by Sanger sequencing. The top 6 genes were NPM1, FLT3-ITD, DNMT3A, CEBPA double allele, IDH1 and IDH2 mutations, and the frequency of each gene in the combining cohort was 36.8%, 19.8%, 20.1%, 5.8%, 14.9% and 22.5%, respectively. In addition, clinical variables such as age, white blood cell counts, genes of IDH1 and DNMT3A mutations, European LeukemiaNet genotype (NPM1 mutations and lacking FLT3-ITD or CEBPA double allele mutations) and treatment protocols were independent factors for predicting the probabilities of overall and event-free survival. The prediction nomograms based on these significant factors showed accurate discrimination. In conclusion, we developed a reliable prediction model for stratifying the risk of elderly patients with CN-AML.