With the improvement of the accuracy and efficiency of laser scanning technology, three-dimensional (3D) terrestrial laser scanning (TLS) technology, which can obtain high precise points-cloud, has been applied in the field of deformation monitoring of subway tunnels, structures building and other fields. However, the data contains many outliers, which are produced by pipe and bolt holes and should be removed for deformation analysis. Based on the 3D TLS technology, taking the subway tunnel in operation as the research object, we propose a new cross section analytical method. We use an ellipse fitting method based on 1-norm minimum residual algorithm to deal with the points which are not on the tunnel cross section. An adaptive threshold selection method is introduced to filter the points which are not on the tunnel cross section. Circular arcs model, instead of overall oval model, is adopted to analyze the deformation of subway tunnel. Deformation analysis of the extraction cross section is tested. The results indicate that this method can eliminate gross errors in the data and the cross section method using circular arcs model can more accurately reflect the tunnel deformation.