The hybrid mechanistic-statistical catchment model SPARROW was applied to predict the mean annual load of nitrogen and phosphorus in streams throughout New Zealand (270,000 km2). The loads from land areas, point sources, and erosion are routed through the drainage network (576,300 reaches) with first-order stream decay and attenuation in lakes and reservoirs. Model parameters were determined by calibration against loads measured in the national water quality network (77 sites). For nitrogen, the model predicted the measured loads well (R2 of 0.956 and RMSE of 0.33 in natural-log space), while for phosphorus the model fit was not as good (R2 of 0.900 and RMSE of 0.58). The predictions of exported yields for streams with catchments > 20 km2 are broadly comparable with previous compilations of yields for various land-use classes for nitrogen, but are larger than the previous measurements for phosphorus. The calibrated stream attenuation and lake/reservoir rates were broadly consistent with previous measurements. The predicted load of total nitrogen (TN) delivered to the coast was 167,700 t yr-1, which is 45% of the loads entering the streams. For total phosphorus (TP) the predicted load to the coast was 63,100 t yr-1, 44% of the load entering the streams. Reservoir/lake attenuation makes a relatively small contribution to the overall attenuation compared with in-stream attenuation (3.5% for nitrogen and 8.5% for phosphorus). The largest contribution of total nitrogen is from pastoral land uses, together accounting for 70% of the total nitrogen load to the coast. Land used for dairying makes a disproportionately large contribution to the load of total nitrogen in relation to the area of land (37% of the load versus 6.8% of the land). For total phosphorus, the highest contribution of the load to the coast is from erosion (53.2%). Point sources contribute only a small proportion of the load to the coast (3.2% for nitrogen, 1.8% for total phosphorus). The monitoring network does not include streams with catchments smaller than 10 km2, so model predictions for streams smaller than 10 km2 should be used with caution. ?? New Zealand Hydrological Society (2005).