The purpose of the present study was the proposal of novel biomarkers in prostate cancer by analyzing mass spectrometry profiles. The latter were obtained from the National Cancer Institute Clinical Proteomics Database. The proposed method applied first a pre-processing pipeline of smoothing, automatic noise estimation, peak detection, and peak alignment, for improving the choice of information reach biomarkers and, second, a two level hierarchical tree structure classification scheme, where at each level a PNN classifier was optimally designed. At the first level, normal cases were discriminated by the PNN from cases with prostate cancer of PSA≥4 and, at the second level, distinction was made by the PNN between cancerous cases with 4≤PSA<10 and PSA>10. Maximum classification accuracies were 97.7% and 95.6% respectively. These high accuracies were achieved by a set of information reach biomarkers, which included the 2068.8m/z, 4675.6 m/z, and 5824.5 m/z values that have been associated with prostate cancer.