Recommendations for phosphorus (P) fertilizer should maximize yields with a minimal level applied. The aim of this work was to improve P fertilizer recommendations for sugarcane in Réunion where there is substantial pedological diversity. As the properties of soils affect the efficiency of P fertilizer, the aim was to adapt the recommended level to the soil type using a new soil classification. We conducted a 2-month pot experiment with sugarcane on 11 different soils from Réunion fertilized with 10 levels of mineral P. Soils were classified according to their mineralogy determined by visible near-infrared spectroscopy (vis-NIRS) analysis. We determined the biomass of the aerial part of the plant, and soil-available P using the Olsen and water extraction methods. Results show that Olsen P was a better predictor of sugarcane response to applied P than P water. P Olsen increased linearly with the level of P added, but the slope changed among soil types, and was higher for halloysite soils than for gibbsite soils and allophane soils. For halloysite soils, sugarcane biomass did not respond to Olsen P. For allophane soils and gibbsite soils, sugarcane biomass increased with Olsen P until a threshold and then became stable. For these soils, the optimal Olsen P value, corresponding to a relative biomass of 95%, was equal to 27 mg/kg, which is close to the values found in the literature in temperate conditions. Based on these results, we determined a method to calculate the optimal level of P fertilizer, considering the soil type determined by NIRS and the Olsen P value measurement. This method should now be integrated into the decision-making tool (Serdaf) developed by the laboratory in Réunion that is intended for use by technicians and farmers involved in sugarcane production.