Field-trial data from a database comprising records of 804 potassium (K) fertiliser trials were used to define the production functions relating exchangeable soil K (quick test K (QTK) 0Á75 mm) to the relative response to fertiliser K applications, for the major soil groups in New Zealand. For all soil groups for which there were sufficient data, the production functions were generally flat in the range QTK 5Á10, and thus the estimated relative pasture production at QTK 5 and QTK 10 were similar. The critical QTK levels to achieve 97% maximum production were relatively well defined, being 6 (5Á8) for sedimentary soils (brown and pallic) and brown soils, and 7 (5Á10) for pumice soils. The data for the allophanic soils were unstable and the best estimate was 6 (5Á10). For the remaining soils groups (podzols and raw soils, organic, recent and gley soils) for which there was much less data, the relationships were essentially flat over the range QTK 2Á10. The probability of pasture responses to applied K increased as soil QTK decreased from 10. For the sedimentary and volcanic soils (including both allophanic and pumice) the probability was about 70Á80% at soil QTK B2. The comparable probabilities were 50Á60% for the recent and gley soils, and 30Á43% for the podzols and raw soils. A feature of the response functions was that some trials were not responsive to fertiliser K despite having low soil QTK. In most cases this could not be attributed to soil K reserves as measured by the soil TBK test (sodium tetra-phenol-boron extractable which measures exchangeable K plus plant-available but non-exchangeable K). Other possible reasons for this feature in the data are discussed, including uptake of K from below the soil sampling depth and the temporal effects of clover responses to applied K. Soil K buffer capacities*the amount of fertiliser K over and above maintenance required to increase soil QTK by 1 unit (DK)*ranged from 50 to 150 kg K ha (1 (average 124) for sedimentary soils. For some soils (developed organic soils, gleyed soils and podzols), fertiliser K had very little effect on QTK (0Á75 mm). It is not clear whether these differences are due to differences in leaching of K from the sampling depth, differences between soils in their ability to absorb and retain applied K or indeed the result of errors in the measurement of this parameter. Estimated maintenance K requirements (i.e. the amount of applied K required to maintain soil QTK levels) increased with increasing soil QTK from 4 to 10, from 0Á150 kg K ha(1 yr (1 to 100Á300 kg K ha (1 yr (1 in situations where losses of K were extreme due to the removal of all harvested clippings. Given the uncertainties in predicting K responses and the amount of fertiliser K required to correct K deficiency, practical suggestions are offered as to how best to diagnose and manage soil K deficiency. Areas for future research to improve the prediction of pasture responses to fertiliser K are also included.