The work aims to propose a potential failure mode and effects analysis method of quality detection in cigarette rolling process based on LS-SVM, so as to solve the problem of low accuracy and efficiency of quality detection in the cigarette rolling process. First of all, FMEA was used to identify the potential failure modes of rolling process and determine the priority of solutions. Secondly, the characteristic signals of key failure modes were obtained by correlation analysis. Finally, the LS-SVM classification model was used to construct process quality detection and diagnosis model. The performance of the proposed method was verified by actual production data. For the identification of seven different failure modes, the overall average identification accuracy of the proposed method was 93.53%, which was much better than BPNN and SVM models in identification accuracy and efficiency and provided a new way for diagnosis of cigarette rolling process.