A new statistical analysis of the in situ scattering phase function measurements performed by the Laboratoire de Météorologie Physique's airborne polar nephelometer is implemented. A principal component analysis along with neural networks leads to the classification of a large data set into three typical averaged scattering phase functions. The cloud classification in terms of particle phase composition (water droplets, mixed-phase, and ice crystals) is done by a neural network and is validated by direct Particle Measuring Systems, Inc., probe measurements. The results show that the measured scattering phase functions carry enough information to accurately retrieve component composition and particle size distributions. For each classified cloud, we support the statement by application of an inversion method using a physical model of light scattering to the average scattering phase function. Furthermore, the retrievals are compared with size composition obtained by independent direct measurements.