Accurate cirque classification is essential for understanding their formation and palaeoclimatic implications. Longitudinal‐profile‐based cirque classification offers advantages over expert classification and parameter‐based methods. This classification fits exponential or power functions to cirque profiles, employing linear classifiers based on the exponential coefficient (c‐value) and cirque height, or a threshold approach based on the power coefficient (b‐value) to classify cirques and non‐cirques. However, previous studies were limited to small sample sets. Our study extends this methodology to more extensive datasets on the southeastern Tibetan Plateau, evaluating its effectiveness across two larger sample sets. Both c‐value and b‐value based methods are tested with two classifiers: the original classifier from previous studies and the parameter‐refitted classifier trained by datasets of this study. The results show that the c‐value‐based method effectively classifies typical cirques and non‐cirques, with notable enhancements in performance based on the refitted classifier compared to the original one. The b‐value‐based method with the refitted classifier performs well in typical cirque identification but is less effective for non‐cirques compared to the original classifier. For all‐type cirques and non‐cirques, both methods demonstrated improved performance in non‐cirque classification although there was a slight trade‐off of cirque classification. Additionally, c‐value based non‐linear classifiers and b‐value optimal threshold for classifying cirque and non‐cirque have been developed, and their improved performance in this classification is discussed. Overall, the longitudinal‐profile‐based classification is more effective for typical cirques and non‐cirques, with potentials for further improvement by considering additional spatial structure information of cirques.