It is now widely recognised that cavitation of the rubber particles plays an important part in the energy-absorbing deformation mechanisms of toughened plastics. Cavitated rubber particles not only reduce the shear yield stress of the polymer, but also act as nuclei for craze initiation [1]. An understanding of the cavitation process is therefore essential in developing theories of rubber toughening. Recent papers have thrown new light on this problem, by using an energy-balance model to relate critical conditions for cavitation to the morphology and properties of the rubber particle [1±4]. The energy for void formation comes from two sources: mechanical loading and differential thermal contraction. Tensile loading produces volume strains in the particle and its surroundings, which are released when the rubber cavitates; additional volume strains of similar magnitude are generated on cooling the polymer from the Tg of the matrix (typically ,100 8C) to 23 8C [5]. One effect of these volume strains on the rubber phase is to increase its ` free volume'' and shift the low-temperature secondary loss peak to lower temperatures [6]. Subsequent cavitation allows the rubber to relax to its equilibrium density, so that the tana maximum shifts back to a temperature near the Tg of the unconstrained elastomer. This principle has been used by Morbitzer and colleagues [7, 8] to demonstrate that cooling ABS polymers to y60 8C is in itself suf®cient to cause cavitation or debonding of selected rubber particles. By varying the particle size and rubber content, they obtained samples in which the volume fractions of cavitated and intact particles were comparable: the low temperature tana peak then split into two, with maxima about 10 K apart. However, splitting of the tana peak in this way is highly unusual; in almost all other cases, a single rubber gives a single loss peak, with no indication of whether the particles are all intact or all cavitated. Most previous studies of rubber particle cavitation have involved mechanical straining, which almost inevitably results in yielding of the matrix, and further expansion of the voids. These complications can be avoided by using cooling alone to induce cavitation; thermal contraction of the rubber particle generates compressive stresses in the neighbouring matrix. From the experimental point of view, the limitations of this method of cavitation are that the voids formed are small (diameters ,50 nm), and occupy ,1% of the volume of the host particle. Detecting voids of this size and concentration using electron microscopy or scattering techniques presents major dif®culties. The aim of the present work is to develop alternative methods, based on dynamic mechanical testing, for detecting cavitation in rubber particles. The basic principle is simple: if the rubber particles are intact, the corresponding tana peak will shift downwards under tensile stress, and upwards under compressive stress. However, if the rubber particles have cavitated, the shift will occur only in compression. Tests were carried out using an Eplexor dynamic mechanical thermal spectrometer (DMTS; Gabo Qualimeter GmbH, Germany). This instrument is able to apply static loads of up to 1500 N to the specimen, in tension or compression, while simultaneously measuring dynamic mechanical properties at small load amplitudes. The material used in this study was a standard high-impact polystyrene (HIPS) containing ,8 wt % of polybutadiene in the form of salami particles with diameters in the range 0.5±5.0 im. In order to vary the mechanical properties of the rubber phase, without affecting the particle morphology or molecular weight of the PS matrix, granules of this HIPS were melt compounded with sulphur at 200 8C in a Werner & P eiderer ZSK-30 twin-screw extruder. This resulted in three blends, containing 0, 0.3 and 0.6% sulphur, which, when dissolved in toluene and centrifuged, respectively gave gel contents of 25.9, 28.3 and 31.9%, and swelling indices of 9.5, 8.6 and 7.3. Granules were compression moulded at 200 8C into 3 mm and 6 mm thick sheets. Rectangular bars measuring 2 mm 3 3 mm 3 18 mm (tensile specimen) and 3 mm 3 3 mm 3 6 mm (compression specimen) were machined from these sheets. Cut surfaces of machined bars were polished with ®ne emery cloth to avoid premature fracture in tension. In both tension and compression tests, static and dynamic stresses were applied parallel to the long direction of the specimen. Specimens were cooled to ,20 K below the Tg of the rubber before applying the static load and starting dynamic testing at 10 Hz. The heating rate was 1 K miny1 and the strain amplitude 0.2%. Each specimen was tested over a range of static stresses, beginning with the lowest. To check reproducibility, three specimens were tested for each condition. Results from these tests are presented in Figs 1 and 2. Two trends are obvious from Fig. 1: (i) crosslinking