A simple, fast and efficient noise-reduction protocol for three-dimensional electron tomographic reconstructions of biological material is presented. The approach is based on iterative application of median filtering and shows promise for automatic noise reduction as a preprocessor for automated data analysis tools which aim at segmentation, feature extraction and pattern recognition. The application of this algorithm produces encouraging results for a wide variety of experimental and synthetic electron tomographic reconstructions. (c) 2006 Elsevier Inc. All rights reserved.