Autophagy is a major catabolic system by which eukaryotic cells undergo self-degradation of damaged, defective, or unwanted intracellular components. An abnormal autophagic level is implicated in the pathogenesis of multiple diseases, including cancers. The aim of this study is to explore the prognostic value of autophagy in bladder cancer (BC), which is a major cause of cancer-related death globally. First, 27 differentially expressed autophagy-related genes (ARGs) were identified in BC patients based on The Cancer Genome Atlas (TCGA) database. Functional enrichment analyses hinted that autophagy may act in a tumor-suppressive role in the initiation of BC. Then, the Cox proportional hazard regression model were employed to identify three key prognostic ARGs (JUN, MYC, and ITGA3), which were related with overall survival (OS) significantly in BC. The three genes represented important clinical significance and prognostic value in BC. Then a prognostic index (PI) was constructed. The PI was constructed based on the three genes, and significantly stratified BC patients into high- and low-risk groups in terms of OS (HR=1.610, 95% CI=1.200-2.160, =0.002). PI remained as an independent prognostic factor in multivariate analyses (HR=2.355, 95% CI=1.483-3.739, <0.001). When integrated with clinical characteristics of age and stage, an autophagy-clinical prognostic index (ACPI) was finally validated, which had improved performance in predicting OS of BC patients (HR=2.669, 95% CI=1.986-3.587, <0.001). The ACPI was confirmed in datasets of GSE13507 (HR=7.389, 95% CI=3.645-14.980, <0.001) and GSE31684 (HR=1.665, 95% CI=0.872-3.179, =0.122). This study provides a potential prognostic signature for predicting prognosis of BC patients and molecular insights of autophagy in BC.