A novel approach to cancer detection in biomarkers spectral subspace (BSS) is proposed. The basis spectra of the subspace spanned by fluorescence spectra of biomarkers are obtained by the Gram-Schmidt method. A support vector machine classifier (SVM) is trained in the subspace. The spectrum of a sample tissue is projected onto and is classified in the subspace. In addition to sensitivity and specificity, the metrics of positive predictivity, Score1, maximum Score1, and accuracy (AC) are employed for performance evaluation. The proposed BSS using SVM is applied to breast cancer detection using four biomarkers: collagen, NADH, flavin, and elastin, with 340-nm excitation. It is found that the BSS SVM outperforms the approach based on multivariate curve resolution (MCR) using SVM and achieves the best performance of principal component analysis (PCA) using SVM among all combinations of PCs. The descent order of efficacy of the four biomarkers in the breast cancer detection of this experiment is collagen, NADH, elastin, and flavin. The advantage of BSS is twofold. First, all diagnostically useful information of biomarkers for cancer detection is retained while dimensionality of data is significantly reduced to obviate the curse of dimensionality. Second, the efficacy of biomarkers in cancer detection can be determined.