It is commonly assumed that soil-15N-labelling provides reliable estimates of N2 fixation in trees by matching N2-fixing and non-N2-fixing tree pairs. As root system is a key parameter in determining suitability of the tree pairs, we compared root architecture of Acacia cyanophylla Lindl. and Casuarina glauca Sieber ex. Spreng. (two N2-fixing trees) with Eucalyptus camaldulensis Dehn. and Ceratonia siliqua L. (two non-N2-fixing trees) at 4-year-old in Mediterranean-semiarid zone. The rhizobium strain used appeared more motile than Frankia strain. A. cyanophylla and E. camaldulensis had extensive rooting area and volume of fine roots, and both species tended to develop marked horizontal rooting, compared to C. glauca and C. siliqua. Characteristics of fine- and horizontal-root components can be used in selecting matched root systems of N2-fixing and reference-paired trees. Root architecture of C. glauca was more similar to C. siliqua, than to E. camaldulensis, and that of A. cyanophylla was more similar to E. camaldulensis than to C. siliqua. Accordingly, E. camaldulensis is an appropriate reference to estimate actual N2 fixation by A. cyanophylla, and C. siliqua is an appropriate reference for C. glauca, when using soil-15N-labelling method in the prevailing site environment.