Atmospheric aerosols consist of a complex heterogeneous mixture of particles. Single-particle analysis techniques are known to provide unique information on the size-resolved chemical composition of aerosols. A scanning electron microscope (SEM) combined with a thin-window energy-dispersive X-ray (EDX) detector enables the morphological and elemental analysis of single particles down to 0.1 microm with a detection limit of 1-10 wt %, low-Z elements included. To obtain data statistically representative of the air masses sampled, a computer-controlled procedure can be implemented in order to run hundreds of single-particle analyses (typically 1000-2000) automatically in a relatively short period of time (generally 4-8 h, depending on the setup and on the particle loading). However, automated particle analysis by SEM-EDX raises two practical challenges: the accuracy of the particle recognition and the reliability of the quantitative analysis, especially for micrometer-sized particles with low atomic number contents. Since low-Z analysis is hampered by the use of traditional polycarbonate membranes, an alternate choice of substrate is a prerequisite. In this work, boron is being studied as a promising material for particle microanalysis. As EDX is generally said to probe a volume of approximately 1 microm3, geometry effects arise from the finite size of microparticles. These particle geometry effects must be corrected by means of a robust concentration calculation procedure. Conventional quantitative methods developed for bulk samples generate elemental concentrations considerably in error when applied to microparticles. A new methodology for particle microanalysis, combining the use of boron as the substrate material and a reverse Monte Carlo quantitative program, was tested on standard particles ranging from 0.25 to 10 microm. We demonstrate that the quantitative determination of low-Z elements in microparticles is achievable and that highly accurate results can be obtained using the automatic data processing described here compared to conventional methods.