Phosphorus (P), a non-renewable resource, needs to be used more efficiently in agriculture. This requires using soil P tests. However, the P test threshold values for fertilizer response depend on many soil properties, some of which may be useful to estimate these threshold values, others not. Therefore, we searched here which soil properties are useful to estimate P threshold values. We calculated the threshold values for Olsen P and 0.01 M CaCl 2 extractable P of 18 representative agricultural soils of the Mediterranean region of Spain. For that, we performed a P starvation experiment in which wheat and sunflower were alternatively pot-cropped. Results show that Olsen P threshold values are negatively correlated to P buffer capacity (r of −0.74, P lower than 0.001), clay content (−0.82, 0.001), pH (−0.76, 0.001), and Fe oxide content (−0.55, 0.05). Multiple regression models involving clay, pH or soil organic C, and phosphatase activity or organic hydrolysable P accounted for as much as 87 % of the variance in calculated Olsen P threshold values. In particular, there is a major effect of organic P on Olsen P threshold values. Single models based on routinely measured soil properties such as clay content and pH made accurate predictions of Olsen P threshold values with r 2 of 0.81 and P lower than 0.001.