This study investigates the strength of a GIS-based and statistical approach for the analysis of stream sediment geochemical data in the development of predictive models for Au and base metal mineralisation in remote and glaciated areas of northern Pakistan. The study is based on a large (similar to 2,000 samples) geochemical dataset generated as a result of extensive sampling campaigns by the Australian Aid Program, Pakistan Mineral Development Corporation (PMDC) and Sarhad Development Authority (SDA). The data were analysed using factor analysis to identify elemental associations and subsequently imported into ArcGIS 9.2 for spatial analysis. The model successfully detected the occurrence of known stibnite veins and disseminations in Chitral, porphyry Cu (skarns) in Domal Nissar and Skardu, and Ni-PGE occurrences in the mafic-ultramafic rocks of Chilas. Spatial catchment anomaly maps (based on 90th, 95th and 99th percentiles) were compared with the GIS-based model to delineate prospective areas, which included Asheriat, Teru, Pakora, Gilgit, Shishi, Imit and Chalt, mainly along the Shyok suture zone, while geochemical anomalies in Kafiristan, upper Mastuj and Lshkarwaz are associated with the Reshun fault. A strong geochemical anomaly in Astore River reflecting mineralization possibly associated with the Indus suture zone. The model was also used to interpret deposit type for gold and base metal anomalies. The majority of Au anomalies were fault related, associated with suture zones, which suggests a dominantly structural rather than lithological control on mineralisation in northern Pakistan.