Accurately segmenting tumors in digital mammography images is a hard task. However, quality of segmentation is important to avoid misdiagnosis. In this work, the GrowCut technique, which is based on cellular automaton, was used to segment tumor regions of digitized mammograms available in the Mini-Mias database. A set of images was submitted to GrowCut technique and segmented images were compared with ground truth in terms of metrics of area, perimeter, Feret's distance, form factor, and solidity. For segmenting tumors, low user interaction is required. Results showed that GrowCut segmentation images obtained similar properties and shape of the ground-truth images, with an average estimated error close to zero, for all metrics analyzed.