LC-MS-based untargeted metabolomics have been proven to be an extremely promising technique to discover biomarkers and explore the mechanisms underlying diseases, which, however, relies heavily on sample pretreatment for metabolite extraction. In the present study, a systematic and pragmatic evaluation of eight protocols employing conventional metabolites extraction strategies, protein precipitation (PPT), and liquid-liquid extraction (LLE), with and without proteinase K (PK) incubation, was performed simultaneously, using human plasma and a mixture of 39 endogenous metabolite standards. These protocols were as follows: (1) PPT with methanol, (2) PPT with acetonitrile, (3) PPT with 2-propanol, (4) two-step LLE of CH2Cl2-MeOH, followed by MeOH-H2O, (5) PK incubation combining two-step LLE of CH2Cl2-MeOH followed by MeOH-H2O, (6) two-step LLE of CHCl3-MeOH, followed by MeOH-H2O, (7) PK incubation combining two-step LLE of CHCl3-MeOH, followed by MeOH-H2O, (8) PK incubation combining MeOH-EtOH PPT. The results suggested that two-step LLE produced broader metabolome coverage than protein precipitation, and the addition of proteinase K enhanced the extraction performance further. Taken together, PK incubation combining two-step LLE of CHCl3-MeOH, followed by MeOH-H2O, was determined to be the most suitable extraction method, because of its broad metabolome coverage, high reproducibility, and satisfactory recovery. Next, the developed optimal sample preparation method was applied successfully to profile the plasma metabolome of colorectal adenoma and uncover its potential mechanism for significant differential changes in linoleic acid and phospholipid metabolism.