Bladder cancer (BLCA) is among the most malignant types of cancer. At present, the prognostic tools available for this disease are insufficient. In the present study, the transcriptome of 1,049 BLCA samples from four datasets from the Gene Expression Omnibus and The Cancer Genome Atlas (TCGA) were analyzed. By utilizing the RNA-seq data provided by TCGA, a risk score staging system model was built to predict the outcome of patients with BLCA using random forest variable hunting and Cox multivariate regression. A total of 7 genes, including zinc finger protein 230, Bcl2-like 14, AHNAK, transmembrane protein 109, apolipoprotein L2, advanced glycation end-product specific receptor and amine oxidase, copper containing 2 were identified as predicting the survival time of patients with BLCA. The patients with a low risk score had a significantly higher survival rate than those with a high-risk score both in the training and validation datasets. Association analyses between risk score and other clinical information were additionally performed; it was demonstrated that the risk score was significantly associated with pathological stage. A nomogram was plotted to compare risk score and other clinical information. The risk score spanned the greatest range of points, indicating the relative accuracy of risk score. In summary, the risk staging model based on the expression of 7 genes is robust and performs more effectively than other clinical information in predicting a prognosis.