OBJECTIVE To determine how good microsatellite analysis (MA) markers in voided urine samples should be to make a surveillance procedure cost-effective in which cystoscopy is partly replaced by MA for patients with non-muscle-invasive urothelial carcinoma (NMI-UC). PATIENTS AND METHODS We constructed a semi-Markov model with a time horizon of 2 years, and a man aged 65 years as reference case. Data were used from a randomized trial (including 448 patients with NMI-UC from 10 hospitals), and from other data sources. The costs and effects (probability of being in a specific health state) were compared for two surveillance strategies: (i) cystoscopy of the urinary bladder every 3 months (conventional arm), and (ii) semi-automated MA of voided urine samples to identify loss of heterozygosity every 3 months, with a control cystoscopy at 3, 12 and 24 months (test arm). Various sensitivity analyses were used to determine the sensitivity, specificity, and costs of MA of urine for which the test arm was as cost-effective as the conventional arm. RESULTS The probability of being without recurrence after 2 years of surveillance was similar (86.6% conventional arm vs 86.3% test arm) with currently available MA markers (sensitivity of 58% and specificity of 73%). However, the test arm led to higher costs ($4104 vs $3433 per head). The test arm would be as effective and cost the same as the conventional arm if the sensitivity of the currently available MA markers was increased at > or =61%, had a specificity of 73%, and decreased the costs of the MA test per follow-up sample from $158 to <$70. CONCLUSIONS Over 2 years, surveillance in which cystoscopy is partly replaced by currently available urinary MA to reduce patient burden can only provide a cost-effective alternative to the conventional surveillance if the MA urine test had a slightly higher sensitivity and its costs could be reduced.