Oxygen saturation (sO2) of RBCs in capillaries can indirectly assess local tissue oxygenation and metabolic function. For example, the altered retinal oxygenation in diabetic retinopathy and local hypoxia during tumor development in cancer are reflected by abnormal sO2 of local capillary networks. However, it is far from clear whether accurate label-free optical oximetry (i.e. measuring hemoglobin sO2) is feasible from dispersed red blood cells (RBCs) at the single-capillary level. The sO2-dependent hemoglobin absorption contrast present in optical scattering signal is complicated by geometry-dependent scattering from RBCs. Here we provide a theoretical model to calculate the backscattering spectra of single RBCs based on the first-order Born approximation, considering the orientation, size variation, and deformation of RBCs. We show that the oscillatory spectral behavior of RBC geometries is smoothed by variations in cell size and orientation, resulting in clear sO2-dependent spectral contrast. In addition, this spectral contrast persists with different deformations of RBCs, allowing the sO2 of individual RBCs in capillaries to be characterized. The theoretical model is verified by Mie theory and experiments using visible light optical coherence tomography (vis-OCT). Thus, this study shows for the first time the feasibility of, and provides a theoretical model for, label-free optical oximetry at the single-capillary level by backscattering-based imaging modalities, challenging the popular view that such measurements are impossible at the single-capillary level. This is promising for in vivo backscattering-based optical oximetry at the single-capillary level, to measure local capillary sO2 for early diagnosis, progression monitoring, and treatment evaluation of diabetic retinopathy and cancer.