Therapeutic conditions for acute leukemia (AL) mainly rely on diagnosis and detection of minimal residual disease (MRD). However, no serum biomarker has been available for clinicians to make diagnosis of AL and assessment of MRD. In this study, we performed bead fractionation/MALDI-TOF-MS analysis on sera from patients with AL. Support vector machine algorithm was used to obtain diagnostic model that discriminated proteomic spectra of patients with AL from that of controls. Twenty-six features with p<0.00001 had optimal discriminatory performance, with 97% sensitivity and 100% specificity. Statistical analysis revealed that two peptides with m/z 1778 and 1865 were gradually decreased in their relative intensities with increase of remission degree. Moreover, the peptide with m/z 1865 was also found to be correlated with AL types. With FT-ICR-MS detection, both the peptides were identified as fragments of complement C3f. Linear regression analysis showed that the combined use of them could discriminate PML/RAR alpha positive M3 from molecular remission M3. Two fragments of complement C3f were significantly correlated with MRD levels and could be used for clinical practice in MRD assessment.