Atmospheric nanoparticles (<0.050 microm) have caused great concern recently due to their potential to affect human health. However, little is known about the chemical composition, sources, and atmospheric behavior of atmospheric nanoparticles. Although gas chromatography/mass spectrometry (GC/MS) after solvent extraction is a commonly used and powerful method for the identification of nonpolar organic compounds in particles, solvent-extraction methods are difficult to apply to nanoparticles because nanoparticles are present in small masses in spite of their high number concentrations. Therefore, we made an attempt to apply thermal desorption-GC/MS (TD-GC/MS), which was expected to be more sensitive than solvent-extraction methods, to atmospheric nanoparticles. A commercial pyrolyzer was used for TD. Prior to the application, the optimum TD-GC/MS conditions for atmospheric particles and diesel exhaust particles (DEP) collected on filters were investigated. Various TD parameters, including desorption time and temperature, were investigated using these test samples and a n-alkanes standard solution. The optimum TD conditions were as follows: ramped desorption from 50 degrees C to 450 degrees C at 50 degrees C min(-1) and then hold for 2 min. Desorption was incomplete at temperatures of 250 degrees C or lower, and considerable pyrolysis occurred at temperatures of 550 degrees C or higher. The TD-GC/MS performance, including the linearity of the calibration curves, repeatability, detection and quantification limits, and sample recovery, under the optimized conditions was evaluated for n-alkanes. It was found that the TD-GC/MS could be applied to extremely small amount of particles (e.g., 5 microg for DEP). The TD-GC/MS was applied to the size-resolved particles, including the nanoparticle fraction (0.0290-0.0580 microm), from roadside atmosphere, and the concentrations of C18-C33 n-alkanes in the particles were determined. The chromatogram pattern of the roadside 0.102-0.163 microm (major size range for DEP) particles was similar to that of the DEP sample. The chromatogram pattern of the roadside nanoparticles was similar to that of diesel lubricating oil, although the proportion of less volatile compounds was slightly larger in the nanoparticles. It is suggested that lubricating oil strongly contributed to the nonpolar organic composition of the roadside nanoparticles, and that more volatile organic compounds in the nanoparticles evaporated in the atmosphere. It was shown that the TD-GC/MS is effective for characterization of atmospheric nanoparticles.