In the human visual system, visual saliency perception is a rapid pre-attention processing mechanism, which can benefit myriad visual tasks, such as segmentation, localization, and detection. While most research is devoted to saliency detection on 2D images and 3D meshes, little work has been performed for efficient saliency detection on 3D point clouds. In this paper, we present a novel point clouds saliency detection method by employing principal component analysis (PCA) in a sigma-set feature space. In this method, we first construct local shape descriptors based on covariance matrices for saliency detection, considering that covariance matrices can naturally model nonlinear correlations of different low-level compact and rotational-invariant features. Secondly, we transform these covariance matrices to vector descriptors in Euclidean vector space by applying the sigma-point technique, which keeps the inner statistics of regions of 3D point clouds. Based on our informative descriptors, PCA is employed in the descriptor space for identifying saliency patterns in a point cloud. Our method shows its advantages of being structure-sensitive, capturing geometry information and computationally efficient. Experimental results demonstrate that our method achieves good performance without using any topological information.