Low application of fertilizer potassium (K) in intensive agriculture leads to a gradual decrease and, afterwards, to a steady state of soil K determined by tests which are based on soil exchangeable K (Kexch) extraction. In this situation, non-exchangeable K (Knon-exch) is being released and therefore real plant availability does not necessarily correspond to soil test K. This incoherency was investigated in 14 agricultural soils with low K as determined by the Mehlich 3 method. Plant-available K released by exhaustive cropping of perennial ryegrass and K determined by the Neubauer seedling test were compared with 8 soil K extraction methods, with the total K content and with the relative content of soil K-bearing minerals. K determined by the ryegrass test (314 mg/kg on average) was more than 3-times higher than K based on Kexch extraction methods, but was from 2 to 3 times lower than K determined by Knon-exch extracting methods. The relative content of mixed-layer phyllosilicates was significantly related to K extracted by soil tests. The relative content of orthoclase correlated only with total K and mica-group minerals with none of the extractions. The best prediction of plant-available K in investigated soils was obtained with sodium tetraphenylboron and StepK methods.