An investigation is presented concerning the ability of neural nets to classify particles using contour data. Different nets were trained to classify limestone, quartz and coffee particles by their outer boundaries. The contour lines of the analysed particles were similar and differed only in a complex way. A new method of interpreting the Fourier coefficients is shown, which might lead to a possibility of defining particle shape classes by examples. Information is given concerning the selection and design of the appropriate neural net, e.g. back-propagation, and self-organizing maps. In addition, a possibility of interpreting the trained neural nets is demonstrated.