An on-line fiber-optic near-infrared spectroscopic method has been developed for real-time monitoring of distillate composition as a reliable way of assessing the degree of a solvent switch process from methanol and n-propanol to isopropyl acetate. Multivariate statistical methods of Multiple Linear Regression and Partial Least Squares were evaluated for quantitative determination of the total alcohol concentrations in the distillate samples. Robust models were obtained by incorporating temperature variance into the calibration set. Reliable prediction of total alcohol concentration was successfully achieved by near-infrared spectroscopy as judged by a comparison with gas chromatography data.