The present research deal with the modeling an analysis of a continuous soft-drinks bottling process by a multivariate approach based on the 3-Way PLS. For this purpose the process behavior was studied, related to a set of off-processing, transitions grades and/or abnormal star-ups, detected in four month process period. A total of 1938 cases were investigated, measuring 8 process variables. Finally, the process was modeled, filtering the high grade of autocorrelation and cross-correlation within the studied variables (particularly T1, T2 and T3). Thereby, the responsible variables for that behavior were detected; thus the effect and grade of disturbance over the process behavior and final product were estimated.