55-nm rubber particles significantly toughened two epoxy systems without loss of Young’s modulus, tensile strength and glass transition temperature. Transmission Electron Microscopy (TEM) showed that the nanoparticles are uniformly dispersed in matrix and have blurred interface with epoxy. 5wt% rubber nanoparticles increased the critical strain energy release rate (G1c) of Jeffamine D230 (J230)-cured epoxy from 175J/m2 to 1710J/m2, while the 10wt% increased G1c of diaminodiphenyl sulfone (DDS)-cured epoxy from 73J/m2 to 696J/m2. This is explained by comparing the surface–surface interparticle distance and total particle surface of nanocomposites with those of composites. The higher the matrix stiffness, the more nanoparticles needed for toughening. Although the 10wt% J230-cured nanocomposite showed a 50% larger size of stress-whitened zone than the 5wt% J230-cured nanocomposite, the 5wt% nanocomposite showed a higher toughness. These nanoparticles were found to pose barriers to the vibration of crosslinked matrix molecules, leading to higher glass transition temperatures. While the matrix shear banding caused by nanoparticle expansion and growth is the major toughening mechanism for the J230-cured nanocomposites, the matrix plastic void growth and deformation are most probably the major mechanisms for the DDS-cured system. Under tensile loading, the nanoparticles in the DDS-cured epoxy created fibrils of 100–200nm in diameter and 3–5μm in length. TEM analysis in front of a subcritically propagated crack tip showed a number of voids of 30–500nm in diameter in the vicinity of the crack, implying that rubber nanoparticles expanded, grew and deformed under loading. Unlike conventional epoxy/rubber composites in which all of the rubber particles in the crack front cavitated under loading, only a portion of the nanoparticles in this study expanded to create voids. Huang and Kinloch’s model developed from composites was found not fit well into these nanocomposites.