Field experiments were conducted over 4 years at 25 sites throughout the main potato-growing areas of South Australia to calibrate the 0.5M sodium bicarbonate extraction procedure as a soil test for potassium and to examine the responses of irrigated potatoes to rates up to 1280 kg/ha K applied as potassium sulfate or potassium chloride either banded at planting or side-dressed after emergence. Potassium application at planting increased yield (P < 0.05) at 10 sites. The mean percentage yield deficit was 21%. There were significant correlations between relative yield and clay and sand contents, cation exchange capacity and bicarbonate-extractable potassium concentration in 0-15-cm surface soil samples collected before potassium fertiliser was applied. Percentage silt and pH were not correlated with relative yield. The bicarbonate-extractable potassium soil test accounted for 70% of the variance in relative yield compared with only 27% for percentage clay and 22% for cation exchange capacity. The prognostic critical bicarbonate-extractable potassium concentrations were: 153 -t 12 mg/kg for the Smith-Dolby bent hyperbola model, 143 mg/kg for the Cate-Nelson separation, and 133 or 176 mg/kg for the Mitscherlich model (concentrations at relative yields of 90 and 95% respectively). Yield responses are likely when soil bicarbonate- extractable potassium concentrations are below 120 mg/kg (deficient range), uncertain between 12 1 and 200 mg/kg (marginal range) and unlikely above 200 mg/kg (non-responsive range). Banding potassium chloride at planting significantly reduced yield compared with potassium sulfate at 2 out of the 14 sites used. The yield reductions occurred with potassium rates of 160 kg/ha or higher. There were significant increases in yield with side-dressing compared with basal application, at 2 of 5 responsive sites using rates of 320 kg/ha K or higher. None of the side-dressing treatments reduced tuber yields below control yields.