BACKGROUND: People with urothelial carcinoma of the bladder are at risk for recurrence and progression following transurethral resection of a bladder tumour (TURBT). Mitomycin C (MMC) and Bacillus Calmette-Guérin (BCG) are commonly used, competing forms of intravesical therapy for intermediate- or high-risk non-muscle invasive (Ta and T1) urothelial bladder cancer but their relative merits are somewhat uncertain. OBJECTIVES: To assess the effects of BCG intravesical therapy compared to MMC intravesical therapy for treating intermediate- and high-risk Ta and T1 bladder cancer in adults. SEARCH METHODS: We performed a systematic literature search in multiple databases (CENTRAL, MEDLINE, Embase, Web of Science, Scopus, LILACS), as well as in two clinical trial registries. We searched reference lists of relevant publications and abstract proceedings. We applied no language restrictions. The latest search was conducted in September 2019. SELECTION CRITERIA: We included randomised controlled trials (RCTs) that compared intravesical BCG with intravesical MMC therapy for non-muscle invasive urothelial bladder cancer. DATA COLLECTION AND ANALYSIS: Two review authors independently screened the literature, extracted data, assessed risk of bias and rated the quality of evidence according to GRADE per outcome. In the meta-analyses, we used the random-effects model. MAIN RESULTS: We identified 12 RCTs comparing BCG versus MMC in participants with intermediate- and high-risk non-muscle invasive bladder tumours (published from 1995 to 2013). In total, 2932 participants were randomised. Time to death from any cause: BCG may make little or no difference on time to death from any cause compared to MMC (hazard ratio (HR) 0.97, 95% confidence interval (CI) 0.79 to 1.20; participants = 1132, studies = 5; 567 participants in the BCG arm and 565 in the MMC arm; low-certainty evidence). This corresponds to 6 fewer deaths (40 fewer to 36 more) per 1000 participants treated with BCG at five years. We downgraded the certainty of the evidence two levels due to study limitations and imprecision. Serious adverse effects: 12/577 participants treated with BCG experienced serious non-fatal adverse effects compared to 4/447 participants in the MMC group. The pooled risk ratio (RR) is 2.31 (95% CI 0.82 to 6.52; participants = 1024, studies = 5; low-certainty evidence). Therefore, BCG may increase the risk for serious adverse effects compared to MMC. This corresponds to nine more serious adverse effects (one fewer to 37 more) with BCG. We downgraded the certainty of the evidence two levels due to study limitations and imprecision. Time to recurrence: BCG may reduce the time to recurrence compared to MMC (HR 0.88, 95% CI 0.71 to 1.09; participants = 2616, studies = 11, 1273 participants in the BCG arm and 1343 in the MMC arm; low-certainty evidence). This corresponds to 41 fewer recurrences (104 fewer to 29 more) with BCG at five years. We downgraded the certainty of the evidence two levels due to study limitations, imprecision and inconsistency. Time to progression: BCG may make little or no difference on time to progression compared to MMC (HR 0.96, 95% CI 0.73 to 1.26; participants = 1622, studies = 6; 804 participants in the BCG arm and 818 in the MMC arm; low-certainty evidence). This corresponds to four fewer progressions (29 fewer to 27 more) with BCG at five years. We downgraded the certainty of the evidence two levels due to study limitations and imprecision. Quality of life: we found very limited data for this outcomes and were unable to estimate an effect size. AUTHORS' CONCLUSIONS: Based on our findings, BCG may reduce the risk of recurrence over time although the Confidence Intervals include the possibility of no difference. It may have no effect on either the risk of progression or risk of death from any cause over time. BCG may cause more serious adverse events although the Confidence Intervals once again include the possibility of no difference. We were unable to determine the impact on quality of life. The certainty of the evidence was consistently low, due to concerns that include possible selection bias, performance bias, given the lack of blinding in these studies, and imprecision.