Over the last decade, major technological advances in X-ray computed tomography (CT) have allowed for the investigation and reconstruction of three-dimensional (3D) natural porous media architectures at very fine scales. Soil scientists can use the internal structure information to develop predictive models for a range of physical, chemical and biological processes in soil. Image segmentation and thresholding are crucial steps when applying these methods to extract complex pore space geometry information from images. Traditional thresholding algorithms face challenges related to the heterogeneity of soil samples, noise and artefacts introduced during the image acquisition process.This study proposes a new segmentation method using local greyscale value (GV) concentration variabilities based on fractal concepts. Singularity maps were created to measure the GV concentration at each point. The C-V method was combined with the singularity map approach (Singularity-CV method) to define thresholds that can be applied to binarize CT images.This study also introduces a new method for creating 3D synthetic soil images based on truncated multifractals that simulate low-contrast and non-bimodal GV histograms. A synthetic soil image was created with the objective to compare traditional segmentation methods (Otsu and maximum entropy) with the Singularity-CV method. We obtained better results in porosity and more amount of pores at all scales than traditional methods, although some small pores were incorrectly identified due to the ability to amplify every anomalous GVs. Misclassification error (ME) was low and similar to Otsu.Two different 3D CT soil images were also used in this analysis, corresponding to samples of the same soil with 1.2 and 1.6gcm−3 bulk densities. After applying the Singularity-CV method to the GV images, the results were compared with the aforementioned traditional segmentation methods. The image comparison was based on the porosity, pore size distribution (PSD) and cumulative pore size distribution (CPSD). The Otsu method achieves a higher porosity than the Singularity-CV method because it defines the largest pores. However, the Singularity-CV method detected more pores at all sizes. The maximum entropy method always yielded the lowest porosity.