To establish the optimum time of radical cystectomy (RC) for patients with recurrent high-risk superficial bladder tumours after the failure of intravesical therapy. Among 318 patients with transitional cell carcinoma treated with RC and with no neoadjuvant therapy, there were 46 with clinical stage Ta, T1 or Tis refractory to transurethral resection associated with intravesical therapy. These patients had at least one of: (i) high-risk superficial bladder tumours after failure of two consecutive induction courses of intravesical therapy; (ii) superficial bladder tumours with prostatic stromal invasion; (iii) superficial bladder tumours with mucosa/ducts involvement after failure of one course of intravesical therapy; (iv) uncontrolled superficial tumours with transurethral resection associated or not with intravesical therapy. Progression and cause-specific survival of these patients were compared to those with muscle-invasive tumours. Univariate and multivariate analyses of predictive factors for cause-specific survival were also used in patients with superficial tumours. The incidence of significant prognostic factors was compared in both superficial and muscle-invasive tumours, as were the progression pattern and survival. The progression-free and cause-specific survival of patients with superficial tumours was 54% and 67%, respectively, with no significant difference from those with muscle-invasive tumours. In multivariate analysis, positive lymph-nodes and prostatic stromal invasion were significant and independent variables for survival. The incidence of positive lymph nodes was 15% vs 30% (P < 0.05) and of stromal invasion was 32% vs 1.5% (P < 0.001) in patients with superficial and muscle-invasive tumours, respectively. Accounting for the progression pattern in patients with superficial tumours, extravesical urothelial recurrence prevailed over local or distant recurrences (30% vs 15%), whereas in patients with muscle-invasive tumours the opposite occurred (5% vs 33%, respectively). The cause-specific survival of patients with superficial tumour and prostatic stromal invasion was one of three, and in those who developed extravesical urothelial recurrence was 28.5%. In patients with recurrent high-risk superficial bladder cancer after intravesical therapy, our criteria for RC were inappropriate, and patients had a survival rate similar to those with muscle-invasive tumours. RC might have been used too late, as there was a high incidence of prostatic stromal invasion and extravesical urothelial recurrence after RC. Both events seem to be responsible of the low cause-specific survival. Predictive factors for progression are needed to indicate early RC in patients with recurrent high-risk superficial tumours. From a previous analysis the pathological pattern of the clinical lack of response (T1, G3, bladder carcinoma in situ and prostate involvement) to intravesical therapy evaluated at 3 months might be important for predicting progression, and an early RC at that time might be useful.