A serum peptide profile contains important bio-information, which may help disease classification. The motivation of this study is to take advantage of porous silicon microparticles with multiple surface chemistries to reduce the loss of peptide information and simplify the sample pretreatment. We developed a multi-dimensional on-particle MALDI-TOF technology to acquire high fidelity and cross-reactive molecular fingerprints for mining disease information. The peptide fingerprint of serum samples from colorectal cancer patients, liver cancer patients and healthy volunteers were measured with this technology. The featured mass spectral peaks can successfully discriminate and predict the multi-category disease. Data visualization for future clinical application was also demonstrated.