Belowground nitrogen (N) transfer may be an important process for the N economy in legume-based intercropping systems. In situ estimates of N transfer based on the natural 15N abundance may be unreliable because 15N discrimination occurs during several soil and plant processes. The aim of this study was to test the use of the stem-15N labeling method to assess in situ N transfer from a legume tree to associated grass. Grass 15N content was measured during a 203-day field experiment from 0 to 5 m from the labeled trees. The results indicated that transfer of the added 15N was limited in space (i.e., up to 1 m from the trees) and delayed in time (i.e., 15N reached the tree roots more than 3 months after labeling), which prevented estimation based on the stem-15N labeling method. Calculations made using the observed natural 15N abundance data suggested that on average 33% of grass N was derived from N2 fixation by the tree. New tree labeling methods have to be developed to verify these values and to obtain reliable estimates of in situ N transfer in agroforestry systems.