Fertiliser recommendation systems should aim at a finer tuning of non-renewable P inputs for agronomic, environmental and economic reasons. Modern decision support systems should take into account the relevant soil characteristics, the P recycling capabilities of the cropping system, and crop requirements for attainable production in a range of soil/weather conditions. Unfortunately, information is still lacking for low input cropping systems in south-western France. In 1968 INRA Toulouse set up a P experiment, which has been going on for 36 years, on a deep alluvial silty-clay/clay soil with varying CaCO3. Four P regimes (P0, P1, P2, P4) were arranged in four blocks with periodic changes in the fertiliser dressings. Wheat, maize, sunflower, sorghum and soybean were tested for grain yield (GY) and grain P concentration (GPC) response to soil Olsen P concentration. The highest GY were observed in both P2 and P4, although P1 yields were significantly lower in only 4 years out of 36. P0 resulted 32 times in lower yields than P2–P4 and 27 times in lower yields than P1. Wheat was the crop most sensitive to the absence of P fertilization (GYP0/GYPmax = 0.72); maize and sorghum were intermediate (0.77) and sunflower was the less sensitive on average (0.83). As the highest GPC values were observed in the P4 treatments, P removal was maximum for P4 (21.9 kg P ha−1 year−1) and minimal in P0 (11.7 kg P ha−1 year−1). The critical soil Olsen P values for yield responses were determined using the Cate–Nelson and Mitscherlich approaches. Although the thresholds differ for the two methods (3.3–7.2 mg P kg−1 with Cate–Nelson; 4.4–11.2 mg P kg−1 with Mitscherlich), crops ranked similarly with both methods.Critical soil P values were lowest for maize and highest for sunflower, while wheat, soybean and sorghum had intermediate values. Because of low-input management and frequent water stress, critical values fall within the lower range of published values. Only in the P4 treatment were P-Olsen values potentially hazardous for the environment (>20 mg P kg−1) 8–10 years after the beginning of the experiment. Annual P dressings of 17.5 kg P ha−1 year−1 (P1) were sufficient to achieve good yields but P dressings of 35 kg P ha−1 year−1 (P2) were necessary to stabilize soil P around the critical level in the calcareous part of the experiment.