Abstract The spatial filtering techniques that are used for the analysis and interpretation of exploration geochemical data to define regional distribution patterns or to outline anomalous areas are, in most cases, based on non-robust statistical methods. The performance of these techniques is heavily influenced by the presence of outliers that commonly exist in the data. This study describes a number of filtering techniques motivated by the development of exploratory data analysis (EDA) and robust statistical procedures. These are the median filter (MF) and the adaptive trimmed mean filter (ATM) for the smoothing of regional geochemical data to reduce spurious variations; two new filters, the fence filter (FF) and the notch filter (NF), have been developed to define geochemical anomalies. The application of the spatial filtering techniques is illustrated by Zn data from approximately 3100 stream sediment samples taken in a regional geochemical survey over 25,000 km 2 of the western margin of the São Francisco Basin, Brazil. Regional distribution patterns for Zn obtained by the MF and ATM filters are clearly related to known stratigraphic units. Anomaly filtering using the FF and NF has delineated most known base metal and gold occurrences, as well as a number of anomalies located in geologically favourable environments but unrelated to any known mineralization. The two anomaly filters have, for the most part, defined the same anomalies in the study area but only the NF highlights the anomaly associated with the important Morro Agudo Pb-Zn deposit, which is too subtle to be immediately apparent in the unprocessed data.