Urine cell-free DNA (cfDNA) is a valuable noninvasive biomarker for cancer mutation detection, infectious disease diagnosis (eg, tuberculosis), organ transplantation monitoring, and prenatal screening. Conventional silica DNA extraction does not efficiently capture urine cfDNA, which is dilute (ng/mL) and highly fragmented [30 to 100 nucleotides (nt)]. The clinical sensitivity of urine cfDNA detection increases with decreasing target length, motivating use of sample preparation methods designed for short fragments. We compared the analytical performance of two published protocols (Wizard resin/guanidinium thiocyanate and Q Sepharose), three commercial kits (Norgen, QIAamp, and MagMAX), and an in-house sequence-specific hybridization capture technique. Dependence on fragment length (25 to 150 nt), performance at low concentrations (10 copies/mL), tolerance to variable urine conditions, and susceptibility to PCR inhibition were characterized. Hybridization capture and Q Sepharose performed best overall (60% to 90% recovery), although Q Sepharose had reduced recovery (<10%) of the shortest 25-nt fragment. Wizard resin/guanidinium thiocyanate recovery was dependent on pH and background DNA concentration and was limited to <35%, even under optimal conditions. The Norgen kit led to consistent PCR inhibition but had high recovery of short fragments. The QIAamp and MagMAX kits had minimal recovery of fragments <150 and <80 nt, respectively. Urine cfDNA extraction methods differ widely in ability to capture short, dilute cfDNA in urine; using suboptimal methods may profoundly impair clinical results.