We demonstrate the first application of Raman spectroscopy in diagnosing nonmalignant, premalignant, malignant, and metastatic stages of breast cancer in a three-dimensional (3-D) cell culture model that closely mimics an in vivo environment. Comprehensive study comparing classification in two-dimensional (2-D) and 3-D cell models was performed using statistical methods composed of principal component analysis for exploratory analysis and outlier removal, partial least squares discriminant analysis, and elastic net regularized regression for classification. Our results show that Raman spectroscopy with an appropriate classification tool has excellent resolution to discriminate the four stages of breast cancer progression, with a near 100% accuracy for both 2-D and 3-D cell models. The diversity in chemical groups related to nucleic acids, proteins, and lipids, among other chemicals, were identified by appropriate peaks in the Raman spectra that correspond to the correct classification of the different stages of tumorigenesis model comprising of MCF10A, MCF10AneoT, MCF10CA1h, and MCF10CA1a cell lines. An explicit relationship between wavenumber and the stages of cancer progression was identified by the elastic net variable selection.