Abstract: To investigate the discrimination methods for arabica green coffee beans processed by different primary methods and to predict their soluble solids content (SSC). A handheld refractometer and Fourier transform infrared spectroscopy (FT-IR) were employed to analyze washed, sun-dried, and honey-processed coffee beans in this study. The discriminant methods and SSC regression prediction models were further developed based on the collected data. The results showed that the SSC content of honey processed coffee beans was the highest (4.86%). Two-dimensional correlation spectroscopy (2D-COS) could accurately identify the difference in spectral features between different samples. The FT-IR spectral data processed by four pretreatment methods, namely Savitzky-Golay smoothing (SG), normalization method (NM), detrending (DT), and multiple scattering correction (MSC), could accurately discriminated different primary processing samples through multivariate statistical analysis. Furthermore, three machine learning models-principal component regression (PCR), partial least squares regression (PLSR), and support vector regression (SVR)—were utilized to predict the SSC of three distinct types of green coffee beans following their primary processing. Notably, the integration of raw data with the PCR model yielded the most accurate predictions, achieving coefficients of determination for calibration (R²_c) and prediction (R²_p) of 0.67 and 0.64, respectively. This study would provide a foundation for evaluating, selecting, enhancing, and improve the quality of coffee beans with different primary processing methods, as well as for perfecting the coffee industry system.