Peak absorbance changes correlated to changes in species concentration have been the norm in applied spectroscopy, while baseline shifts have been more of an inconvenience. Taking the first or second derivative of the spectra eliminates these baseline shifts. However, with multivariate techniques becoming more readily available, repeatable baseline changes may now be monitored and correlated to specific physical changes. This concept has been studied by monitoring the concentration of titanium dioxide (TiO2), a white inorganic filler, in molten poly(ethylene terepthalate) (PET). Various mixtures of filled and unfilled PET resins were run through a single-screw extruder, and near infrared spectra were collected in-line by using a flow cell, housing two fiber-optic probes, and mounted downstream of the extruder. The presence of titanium dioxide caused the scattering of light that resulted in a systematic baseline shift. The baseline shifts were correlated to the TiO2 concentration data. Multivariate techniques involving the use of singular value decomposition (SVD) to perform partial least squares regression (PLS) were applied to quantitatively determine TiO2 content in the PET melt stream. Standard error of prediction (SEP) values of about 1% were obtained for a model based on two factors.