Qualitative trend analysis (QTA) is an effective tool for process data analysis, the applications of which can be found in a variety of fields, such as process monitoring, fault diagnosis, and data mining. Reliable and accurate trend extraction of sensor data is the first and indispensable step in QTA. In this article, a new trend extraction algorithm is developed that is based on global optimization of the polynomial fit of the process data. Different from most existing works, this newly proposed algorithm solves the trend extraction task by simultaneously and globally estimating the episode number, the boundary time points of the episode, and the fitted polynomial coefficients, which shows improved performance over other nonglobally optimal trend extraction algorithms and requires less a priori knowledge than the existing globally optimal trend algorithms. The effectiveness of the algorithm is illustrated by testing on a variety of simulation and real blast furnace data. © 2017 American Institute of Chemical Engineers AIChE J, 63: 3374–3383, 2017