Within the framework of routine application of image analysis in a pharmaceutical laboratory, a compromise should be established between the number of particles, the analysis time and the accuracy of the results. In the case of nonacicular particles, 500 particles constitute the minimal set to be analysed to get basic information on size and shape distributions. The best choice is 1000 to 1500 particles. As complex shapes should be described by a set of parameters, statistical data mining techniques allow to globally compare samples from the point of view of their size and morphology. For acicular particles, frequently seen in pharmaceutical powders, an algorithm based on the Hough transform and the Euclidian Distance Map transform has been developed: it allows to estimate the length and the breadth of the particles, even when they are in (limited) contact with one another.