Short-term legume–cereal rotation systems (referred to as improved fallows) with N2 fixing leguminous species are being actively promoted to improve soil fertility in fallowed fields of smallholder farms in many parts of the tropics. Few estimates of N2-fixation in deep-rooted woody fallow species are available due to methodological difficulties. We evaluated and developed the natural 15N abundance method for assessing N2-fixation in herbaceous and woody legumes on a Kandiudalfic Eutrudox in western Kenya by (i) assessing isotopic discrimination during N2-fixation and translocation, (ii) measuring variability of 15N with depth, (iii) comparing with an independent method (ureide assay) and (iv) using several non-fixing reference plants. Most tested tree/shrub legumes showed no 15N discrimination during N2-fixation (i.e. whole plant 15N was close to 0). Significant 15N isotopic discrimination occurred during translocation of fixed N, which resulted in 15N depletion in shoots (up to -1.76 in Sesbania sesban) compared with roots and nodules which were 15N enriched. Soils were highly enriched in 15N (8.2–10.8) with little variation with depth to 2 m. 15N signatures of plant available N measured using non-fixing reference plants were lower than those of total soil N. 15N of the non-fixing reference species maize, Lantana camara and Tithonia diversifolia varied by 2.0 and resulted in corresponding variation of N2-fixation estimates for respective species. 15N based estimates of N2-fixation of pigeonpea and siratro were linearly related with those obtained using the ureide method (R2=0.80, slope=0.82) and confirmed the utility of the 15N natural abundance method. Field observations showed that under non-PK limiting growth conditions, the proportion of N2 fixed ranged 75–83, 63–74, 55–67, 46–59, 36–54, 35–50, and 36–51 or Crotalaria grahamiana, Tephrosia vogelii, pigeonpea (Cajanus cajan), S. sesban, Calliandra calothyrsus, siratro (Macroptilium atropurpureum) and groundnut (Arachis hypogaea). This resulted in average amounts of N2 fixed of 142, 100, 91, 52, 24, 64 and 8 kg N ha-1, respectively, 9 months after planting. The amount of soil derived N ranged between 31 and 57 kg N ha-1 in woody species. The net N balance of woody fallows (after adjusting for N export in wood) was highest in Crotalaria due to high N2-fixation and small amount of N exported in wood. Overall, partial N balances indicated that additional N derived from N2-fixation constituted a major component of recyclable N of the system. We conclude that, in soil with sufficient and relatively uniform background 15N abundance and using appropriate, or a range of, non-fixing reference plants, the natural 15N abundance method is a useful tool for estimating the amount of N derived from N2-fixation by field grown herbaceous and woody legumes.