Abstract. Although small particles (size between 25 lm and 200 lm) are frequently observed within ice and water clouds, they are not generally used properly for the calculation of structural, optical and microphysical quantities. Actually neither the exact shape nor the phase (ice or water) of these particles is well de®ned since the existing pattern recognition algorithms are only e cient for larger particle sizes. The present study describes a statistical analysis concerning small hexagonal columns and spherical particles sampled with a PMS-2DC probe, and the corresponding images are classi®ed according to the occurrence probability of various pixels arrangements. This approach was ®rst applied to synthetic data generated with a numerical model, including the e ects of di raction at a short distance, and then validated against actual data sets obtained from in-cloud¯ights during the pre-ICE'89 campaign. Our method allows us to di erentiate small hexagonal columns from spherical particles, thus making possible the characterization of the three dimensional shape (and consequently evaluation of the volume) of the particles, and ®nally to compute e.g., the liquid or the ice water content.