Label-free　surface-enhanced　Raman　spectroscopy　(SERS)　is　capable　of　capturing　rich　compositional　information　from　complex　biosamples　by　providing　vibrational　spectra　that　are　crucial　for　biosample　identification.　However,　increasing　complexity　and　subtle　variations　in　biological　media　can　diminish　the　discrimination　accuracy　of　traditional　SERS　excited　by　a　single　laser　wavelength.　Herein,　we　introduce　a　multiwavelength　SERS　approach　combined　with　machine　learning　(ML)-based　classification　to　improve　the　discrimination　accuracy　of　human　urine　specimens　for　bladder　cancer　(BCa)　diagnosis.　This　strategy　leverages　the　excitation-wavelength-dependent　SERS　spectral　profiles　of　complex　matrices,　which　are　mainly　attributed　to　wavelength-related　vibrational　changes　in　individual　analytes　and　differences　in　the　variation　ratios　of　SERS　intensity　across　different　wavelengths　among　various　analytes.　By　capturing　SERS　fingerprints　under　multiple　excitation　wavelengths,　we　can　acquire　more　comprehensive　and　unique　chemical　information　on　complex　samples.　Further　experimental　examinations　with　clinical　urine　specimens,　supported　by　ML　algorithms,　demonstrate　the　effectiveness　of　this　multiwavelength　strategy　and　improve　the　diagnostic　accuracy　of　BCa　and　staging　of　its　invasion　with　SERS　spectra　from　increasing　numbers　of　wavelengths.　The　multiwavelength　SERS　holds　promise　as　a　convenient,　cost-effective,　and　broadly　applicable　technique　for　the　precise　identification　of　complex　matrices　and　diagnosis　of　diseases　based　on　body　fluids.