A large-scale column study was conducted in Melbourne, Australia, to test the performance of stormwater biofilters for the removal of sediment, nitrogen and phosphorus. The aim of the study was to provide guidance on the optimal design for reliable treatment performance. A variety of factors were tested, using 125 large columns: plant species, filter media, filter depth, filter area and pollutant inflow concentration. The results demonstrate that vegetation selection is critical to performance for nitrogen removal (e.g. Carex appressa and Melaleuca ericifolia performed significantly better than other tested species). Whilst phosphorus removal was consistently very high (typically around 85%), biofilter soil media with added organic matter reduced the phosphorus treatment effectiveness. Biofilters built according to observed 'optimal specifications' can reliably remove both nutrients (up to 70% for nitrogen and 85% for phosphorus) and suspended solids (consistently over 95%). The optimally designed biofilter is at least 2% of its catchment area and possesses a sandy loam filter media, planted with C. appressa or M. ericifolia. Further trials will be required to test a wider range of vegetation, and to examine performance over the longer term. Future work will also examine biofilter effectiveness for treatment of heavy metals and pathogens.