The diagnosis of bladder cancer (BC) is currently based on cystoscopy, which is invasive and expensive. Here, we describe a noninvasive profiling method for carbonyl metabolic fingerprints in BC, which is based on a desorption, separation, and ionization mass spectrometry (DSI-MS) platform with N,N-dimethylethylenediamine (DMED) as a differential labeling reagent. The DSI-MS platform avoids the interferences from intra- and/or intersamples. Additionally, the DMED derivatization increases detection sensitivity and distinguishes carboxyl, aldehyde, and ketone groups in untreated urine samples. Carbonyl metabolic fingerprints of urine from 41 BC patients and 41 controls were portrayed and 9 potential biomarkers were identified. The mechanisms of the regulations of these biomarkers have been tentatively discussed. A logistic regression (LR) machine learning algorithm was applied to discriminate BC from controls, and an accuracy of 85% was achieved. We believe that the method proposed here may pave the way toward the point-of-care diagnosis of BC in a patient-friendly manner.