Bladder cancers can be categorized into subtypes according to gene expression patterns. P53-like muscle-invasive bladder cancers are generally resistant to cisplatin-based chemotherapy, but exhibit heterogeneous clinical outcomes with a prognosis intermediate to that of the luminal and basal subtypes. The optimal approach to p53-like tumors remains poorly defined and better means to risk-stratify such tumors and identification of novel therapeutic targets is urgently needed. MicroRNAs (miRNAs) play a key role in cancer, both in tumorigenesis and tumor progression. In the past few years, miRNA expression signatures have been reported as prognostic biomarkers in different tumor types including bladder cancer. However, miRNA’s expression does not always correlate well with its activity. We previously developed ActMiR, a computational method for explicitly inferring miRNA activities. We applied ActMiR to The Cancer Genome Atlas (TCGA) bladder cancer data set and identified the activities of miR-106b-5p and miR-532-3p as potential prognostic markers of the p53-like subtype, and validated them in three independent bladder cancer data sets. Especially, higher miR-106b-5p activity was consistently associated with better survival in these data sets. Furthermore, we experimentally validated causal relationships between miR-106-5p and its predicted target genes in p53-like cell line HT1197. HT1197 cells treated with the miR-106b-5p-specific inhibitor were more invasive while cells treated with the miR-106b-5p-specific mimic were less invasive than corresponding controls. Altogether, our results suggest that miR-106b-5p activity can categorize p53-like bladder tumors into more and less-favorable prognostic groups, which provides critical information for personalizing treatment option for p53-like bladder cancers.