Soil organic carbon (SOC) is a crucial component of the global carbon cycle, playing a vital role in maintaining soil health, promoting plant growth, and regulating climate change. Owing to its efficiency and non-destructive nature, spectroscopic analysis has become an important method for rapid SOC determination. However, the mechanisms by which external environmental factors (e.g., illumination, atmospheric conditions) and sensor parameters (e.g., spectral and spatial resolution) influence SOC retrieval accuracy across different spectral data sources remain insufficiently understood. In this study, conducted in northeastern Inner Mongolia Autonomous Region, we collected 160 surface (0-20 cm) SOC samples and concurrently acquired proximal hyperspectral measurements (under indoor artificial illumination and outdoor sunlight), as well as spaceborne multispectral (Landsat-8, Sentinel-2) and hyperspectral (ZY1-02D) data. Random Forest (RF) and Support Vector Machine (SVM) algorithms were used to construct SOC retrieval models for each data source. By systematically comparing model performance, we analyzed how environmental factors and sensor parameters affect SOC retrieval accuracy. The results show that: (1) due to stable and controllable spectral signals, indoor artificial illumination yields slightly higher retrieval accuracy than outdoor sunlight, though the difference is small, indicating that natural light variability has a limited impact on SOC retrieval; (2) proximal hyperspectral data achieve significantly higher retrieval accuracy than satellite multispectral and hyperspectral data, primarily because satellite observations are affected by atmospheric scattering, water-vapor absorption, and mixed pixels; and (3) among satellite data, the hyperspectral ZY1-02D provides higher retrieval accuracy than multispectral sensors, while the improvement from Sentinel-2’s higher spatial resolution (10 m) over Landsat-8 (30 m) is limited, suggesting that spectral resolution contributes more to SOC retrieval accuracy than spatial resolution. Through a comprehensive comparison of multi-source spectral data combined with machine learning algorithms, this study elucidates the influence of environmental factors and sensor parameters on SOC retrieval accuracy and offers practical guidance for data selection in regional-scale SOC estimation and mapping.