Biofilms are microbial aggregates of microorganisms surrounded by a hydrogel-like matrix formed by extracellular polymeric substances (EPS). The formation of biofilms is intrinsically complex, from the attachment of microbial cells to the dispersion of the biofilm. Meanwhile, the three-dimensional framework built up by EPS changes with time and protects the microorganisms against environmental stress. Simultaneous acquiring chemical and structural information within the biofilm matrix is vital for the cognition and regulation of biofilms, yet it remains a great challenge due to the sample complexity and the limited approaches. In this study, confocal Raman microscopy and non-negative matrix factorization (NMF) analysis were combined to investigate spatiotemporal organization of Escherichia coli biofilms during development at molecular-level detail. The alternating non-negative least square (ANLS) approach was incorporated with the sequential coordinate-wise descent (SCD) algorithm to realize the NMF analysis for the large-scale hyperspectral dataset. As a result, three components, including bacteria, protein and polyhydroxybutyrate (PHB) were successfully resolved from the spectra of E. coli biofilm. Furthermore, the structural changes of biofilms could be visualized and quantified by their abundances derived from the NMF analysis, which might be related to the nutrient and oxygen gradient and physiological functions. This methodology provides a comprehensive understanding of the chemical constituents and their spatiotemporal distribution within the biofilm matrix. Furthermore, it also shows great potential for the analysis of unknown and complex biological samples with 3D Raman mapping.