Mass spectrometry (MS) is a technology recently used for high dimensionality detection of proteins in proteomics. However, due to the high resolution and noise of MS data (MALDI-TOF), almost existing MS analysis algorithms are not robust with noise and run slowly. Developing new ones is necessary to analyze such data. In this paper, we propose a novel feature extraction method considering the inherent noise of mass spectra. The proposed method combines stationary wavelet transformation (SWT) and bivariate shrinkage estimator for MS feature extraction and denoising. Then, statistical feature testing is applied to denoised wavelet coefficients to select significant features used for biomarker identification. To evaluate the effectiveness of proposed method, a double cross-validation support vector machine classifier, which has high generalizability, and a fast Modest AdaBoost classifier, which improves significantly experimental runtime, are applied for cancer classification based on selected features by proposed method. Several experiments are carried out to evaluate the performance of our proposed methods. The results show that our proposed method can be an effective tool for analyzing MS data.