Background Bladder cancer displays a broad mutational spectrum and intratumor heterogeneity (ITH), which results in difference in molecular phenotypes and resistance to therapies. However, there are currently no clinically available measures to predict patient prognosis using ITH. We aimed to establish a clinically relevant biomarker by using ITH for informing predictive of outcomes. Methods We used the Bioconductor R package Maftools to efficiently and comprehensively analyze somatic variants of muscle-invasive bladder cancer (MIBC) from The Cancer Genome Atlas (TCGA). We then used a mutant-allele tumor heterogeneity (MATH) algorithm to measure ITH and explored its correlation with clinical parameters as well as mutational subtypes. Results We observed a broad range of somatic mutations in MIBC from TCGA. MATH value was higher for the high-grade group than for the low-grade group (p < 0.05). There was a strong correlation between higher MATH value and presence of TP53 mutations (p = 0.008), as well as between lower MATH value and presence of FGFR3 mutations (p = 0.006). Patients with FGFR3 mutation and low MATH value exhibit longer overall survival time than that of all BLCA patients (p = 0.044), which was replicated in another bladder cancer database composed of 109 BLCA patients. Conclusion Measures of tumor heterogeneity may be useful biomarkers for identifying patients with bladder cancer. Low MATH value was an independent risk factor that predicted better prognosis for patients with FGFR3 mutation compared to all BLCA patients.