In order to correctly extract the geometric features of anatomies in CT images, we proposed an approach of edge detection for CT images based on multiscale analysis. The analysis for multiscale edge detection was implemented by using the negative derivatives of a smoothing function containing a scale factor as the wavelets to perform wavelet transform on CT images, and by locating the local modulus maxima of the transform to depict the geometric features of anatomies in the images. We also proposed a simple method for identifying the local modulus maxima points. In addition, in order to address the relatively large CT image noise, we utilized the thresholds, that were obtained by multiplying the means square root of all local modulus maxima with a factor related to the scale, and achieved clean edges consequently. The resulting pictures of local modulus maxima after the thresholding showed that the edge information of different objects of different sizes could be revealed by different edge detections under different scales. It is indicated that the method proposed can extract the geometric contour features of the anatomies of interest correctly after suppressing the noise effectively.