A rapid and non-intrusive on-line NIR imaging sensor is developed for monitoring spatio-temporal crystallinity variations across the surface of polymer films. A multivariate image analysis and regression (MIA/MIR) approach is proposed and compared with standard NIR calibration techniques using averaged spectra or second order derivatives combined with PLS regression. Predictions of both the local and global crystallinity variations of HDPE, LDPE, and PIP polymer samples were obtained with each approach. Our results show that small variations in crystallinity introduced by changes in cooling rates can be predicted within experimental errors. Crystallinity spatial distributions were also validated and found consistent with processing conditions.