Understanding the composition of bone tissue at the submicron level is crucial to elucidate factors contributing to bone disease and fragility. Here, we introduce a novel approach utilizing optical photothermal infrared (O-PTIR) spectroscopy and imaging coupled with machine learning analysis to assess bone tissue composition at 500 nm spatial resolution. This approach was used to evaluate thick bone samples embedded in typical poly(methyl methacrylate) (PMMA) blocks, eliminating the need for cumbersome thin sectioning. We demonstrate the utility of O-PTIR imaging to assess the distribution of bone tissue mineral and protein, as well as to explore the structure-composition relationship surrounding microporosity at a spatial resolution unattainable by conventional infrared imaging modalities. Using bone samples from wildtype (WT) mice and from a mouse model of osteogenesis imperfecta (OIM), we further showcase the application of O-PTIR spectroscopy to quantify mineral content, crystallinity, and carbonate content in spatially defined regions across the cortical bone. Notably, we show that machine learning analysis using support vector machine (SVM) was successful in identifying bone phenotypes (typical in WT, fragile in OIM) based on input of spectral data, with over 86% of samples correctly identified when using the collagen spectral range. Our findings highlight the potential of O-PTIR spectroscopy and imaging as valuable tools for exploring bone submicron composition.