Ovarian cancer is one of the most aggressive and heterogeneous female tumors in the world, and serous ovarian cancer (SOC) is of particular concern for being the leading cause of ovarian cancer death. Due to its clinical and biological complexities, ovarian cancer is still considered one of the most difficult tumors to diagnose and manage. In this study, three datasets were assembled, including 30 cases of serous cystadenoma (SCA), 30 cases of serous borderline tumor (SBT), and 45 cases of serous adenocarcinoma (SAC). Mueller matrix microscopy is used to obtain the polarimetry basis parameters (PBPs) of each case, combined with a machine learning (ML) model to derive the polarimetry feature parameters (PFPs) for distinguishing serous ovarian tumor (SOT). The correlation between the mean values of PBPs and the clinicopathological features of serous ovarian cancer was analyzed. The accuracies of PFPs obtained from three types of SOT for identifying dichotomous groups (SCA versus SAC, SCA versus SBT, and SBT versus SAC) were 0.91, 0.92, and 0.8, respectively. The accuracy of PFP for identifying triadic groups (SCA versus SBT versus SAC) was 0.75. Correlation analysis between PBPs and the clinicopathological features of SOC was performed. There were correlations between some PBPs ($\delta$, $\beta$, $q_L$, $E_2$, rq_cross, $P_2$, $P_3$, $P_4$, and $P_5$) and clinicopathological features, including the International Federation of Gynecology and Obstetrics (FIGO) stage, pathological grading, preoperative ascites, malignant ascites, and peritoneal implantation. The research showed that PFPs extracted from polarization images have potential applications in quantitatively differentiating the SOTs. These polarimetry basis parameters related to the clinicopathological features of SOC can be used as prognostic factors.