Nitrogen-15 (15N) fractionation during N mineralisation is frequently invoked to interpret data of N cycling in N-isotope-based ecosystem studies. Because direct measurement of actual fractionation (Dact) is difficult to perform, it is currently estimated from laboratory or field incubation experiments (apparent fractionation, Dapp). However, the procedure of calculation may induce bias in the estimations due to several factors controlled by litter inputs. The objective of this study was to assess the effects of litter inputs and the size and 15N content of organic N fractions on Dapp. Laboratory incubations were performed using soils receiving different amount of root litter from legume trees and grass in an agroforestry plot. The 15N contents of organic N and mineralised N were determined. The results were interpreted with a ‘three-N’ pool model considering two labile organic N pools derived from litter of the tree and the grass, and a recalcitrant organic N pool comprising organic N in mineral soil. The model was used to explore mechanisms that could explain the Dapp obtained in the incubation experiment. Model simulations well duplicated the observed relationships between organic N, mineralised N, their 15N content, and Dapp. Dapp varied from 1.2‰ to 2.4‰. It was negatively correlated with the distance from trees (R2 = 0.91, P < 0.05) and positively correlated with total organic N (R2 = 0.81, P < 0.05). Model simulations indicated that these correlations did not reflect a control of 15N fractionation, but were an artefact induced by the higher decomposition rate of the labile organic N depleted in 15N. This induced an overestimation of 15N fractionation that varied from 10% to 55%. Our results indicated that the procedure currently applied to estimate 15N fractionation during mineralisation may induce large overestimations depending upon the amount of litter input, and the size and 15N content of the organic N pools.