Endometrial diseases, including endometrial polyps (EP), endometrial cancer (EC) and endometrial hyperplasia (EH), are common gynecological diseases that affect women of childbearing and perimenopausal age. Clinically, biopsy or imaging methods are usually used to screen and diagnose these diseases; however, due to the invasiveness and heterogeneity of these tests, a noninvasive, convenient, objective and accurate biomarker is needed for the differential diagnosis of EP, EC or EH. In the present study, serum samples from 326 patients with endometrial diseases and 225 healthy volunteers were analyzed using nontargeted lipidomics. A combination of multivariate and univariate analyses was used to identify and qualify six, eight and seven potential biomarkers in the sera from patients with EP, EC and EH, respectively. Using a logistic regression algorithm and receiver operating characteristic (ROC) curve analysis, a biomarker panel including four specific EP biomarkers, 6‐keto‐PGF1α, PA(37:4), LysoPC(20:1) and PS(36:0), showed good classification and diagnostic ability in distinguishing EP from EC or EH. The biomarker panel for distinguishing EP from EC yielded an area under the curve (AUC) of 0.915, sensitivity of 100% and specificity of 72.41%, while that for distinguishing EP from EH yielded an AUC of 1.000, sensitivity of 100% and specificity of 100%. The two diagnostic models also showed good diagnostic abilities in the validation set. Therefore, this biomarker panel can be used as a rapid diagnostic method to assist in imaging examinations and provide a reference for clinicians in the identification and diagnosis of endometrial diseases.