To compare laryngeal carcinoma tissues and pericarcinous tissues in Raman spectrogram to obtain their respective features and patterns, we establish a diagnose model for laryngeal carcinoma. A total of 110 samples including 54 laryngeal carcinoma samples and 56 pericarcinous samples are acquired from patients undergoing laryngectomy from October 2014 to August 2015 in otolaryngology, head & neck surgery of Xiangya hospital of central south university. All of the samples are processed by the accordant method before Roman spectrum detection. Original data are collected for pretreatment and statistical analysis. The results show that the average Roman spectrum intensity of the laryngeal carcinoma tissues is greater than that of the pericarcinous tissues, and the laryngeal carcinoma tissues have stable Roman peak groups. Statistical difference of the two groups of samples are verified through student′s t-test in bands with wave numbers of 150-1859 cm-1, 1864-1872cm-1, 1890-1898.5 cm-1, 1900-1924 cm-1, 1964-1968 cm-1, 1993-1998.5 cm-1, and 2010.5-3476 cm-1(significance level p＜0.05). Three principal components are extracted with the help of analysis of the significant wave numbers using principal components analysis-linear discriminant analysis (PCA-LDA). Using them as inputs of LDA, we calculate discriminant function coefficient and establish a discriminant model. The discriminant model has better specificity of 80.4% and sensitivity of 87.0% for sample prediction. The laryngeal carcinoma diagnose model is evaluated by received operator characteristic (ROC) curve method, and the value of area under the curve (AUC) is 0.877. There are remarkable and stable differences between the laryngeal carcinoma tissues and the pericarcinous tissues. The discriminant model established by the method of multivariate statistical analysis has better reliability. This study demonstrates that Raman spectroscopy has the potential for the noninvasive, real-time diagnosis and detection of laryngeal carcinoma at the molecular level.