Abstract: The stability of tunnel surrounding rock is affected by many factors, and there are complex nonlinear relationships among various influencing factors. In order to accurately determine the surrounding rock stability category, five factors including rock quality index RQD, rock uniaxial saturated compressive strength Rw, integrity coefficient Kv, structural plane strength coefficient Kf and groundwater seepage ω　are selected as the main factors affecting the stability of surrounding rock, through principal component analysis (PCA), we reduce the above influencing factors into two independent principal components, and use the relevance vector machine (RVM) to establish a nonlinear mapping relationship between the principal component variables and the surrounding rock stability categories. The results of the evaluation are entered into a “one-against-one” voter to determine the stability category. The PCA-RVM surrounding rock stability identification model is applied to actual projects, and 21 sets of samples are used to predict 6 sets of samples, of which 5 sets of prediction results are consistent with the actual category. The results show that the proposed recognition model is suitable for small sample problems. By setting an appropriate bandwidth and iteration times, the accuracy of the model can be effectively improved and the false positive rate can be reduced.