Bladder cancer is diagnosed by cystoscopy, a costly and invasive procedure that is associated with patient discomfort. Analysis of tumor-specific markers in DNA from sediments of voided urine has the potential for non-invasive detection of bladder cancer; however, the sensitivity is limited by low fractions and small numbers of tumor cells exfoliated into the urine from low-grade tumors. The purpose of this study was to improve the sensitivity for non-invasive detection of bladder cancer by size-based capture and enrichment of tumor cells in urine. In a split-sample set-up, urine from a consecutive series of patients with primary or recurrent bladder tumors (N = 189) was processed by microfiltration using a membrane filter with a defined pore-size, and sedimentation by centrifugation, respectively. DNA from the samples was analyzed for seven bladder tumor-associated methylation markers using MethyLight and pyrosequencing assays. The fraction of tumor-derived DNA was higher in the filter samples than in the corresponding sediments for all markers (p<0.000001). Across all tumor stages, the number of cases positive for one or more markers was 87% in filter samples compared to 80% in the corresponding sediments. The largest increase in sensitivity was achieved in low-grade Ta tumors, with 82 out of 98 cases positive in the filter samples (84%) versus 74 out of 98 in the sediments (75%). Our results show that pre-analytic processing of voided urine by size-based filtration can increase the sensitivity for DNA-based detection of bladder cancer.