Phosphorus (P) is the primary limiting nutrient for crop production in highly weathered tropical soils. The deficiency is mainly caused by strong adsorption of H2PO4¿ to Al- and Fe-(hydr)oxides, which turns large proportions of total P into a form that is unavailable to plants. Soil management modifies P dynamics. Some plants, including trees used in agroforestry systems, are known to accelerate P cycling. The objective of this paper was to use phosphorus 31 nuclear magnetic resonance (31PNMR) to evaluate the inorganic (Pi) and organic P (Po) compounds in Oxisol from two agroforestry (15 and 19 years old) and two conventional (full-sun, monoculture, ca. 15¿20 and 20¿24 years) coffee systems at three different depths (2¿3, 10¿15 and 40¿60 cm). We hypothesised that the amounts of (1) organic P and (2) diester are higher in agroforestry fields than in conventional coffee fields and (3) the organic P and the diester decrease less with depth in the agroforestry systems than in the conventional systems. The soils were sampled from on-farm experiments in the Atlantic Coastal Rainforest, Brazil. The soil P was extracted with NaOH 0.5 M+EDTA 0.1 M. Resin chelex-X100 was used to remove the paramagnetic ions. The total P in the NaOH¿EDTA extract was measured through ICP and the Pi by the ammonium molybdate¿ascorbic acid method. Po was calculated as the difference between total P and Pi. The amount of Po was higher, the decrease of Po with depth was more sharp and the Po/total P was lower in the conventional systems than in the agroforestry systems. Based on literature and standards, 31PNMR signals were interpreted as inorganic orthophosphate, orthophosphate monoester (inositol phosphates and mononucleotides), orthophosphate diester (phospholipids, nucleic acids and teichoic acid) and pyrophosphates. The proportion of organic P (Po) was on average 47%, consisting of monoester (95%) and diester (5%). The amounts of diester phosphates did not differ between systems, but the proportion of diester to total spectra areas was higher and the decrease of diester with depth was less in the agroforestry than in the conventional systems. The proportions of inorganic P to total P consisted on average of 45% orthophosphate and 8% pyrophosphate. Our results suggest that agroforestry systems influence the dynamics of P through the conversion of part of the inorganic P into organic P. The effect was higher in deeper layers. Because the rate of cycling is higher for organic P than for inorganic P and for diester than for monoester, and because the P in deep layers is normally less available to crop plants, agroforestry would maintain larger fractions of P available to agricultural crops, thereby reducing P losses to the unavailable pools. The rate and the impacts of these changes on P cycling and efficiency of P use of the crops in the long-term need to be further examined and understood, for full evaluation of the importance of agroforestry in soil P utilisation.