The reconstruction of particle size distributions from chord length distributions is well known to be a tricky task. The usual approach is to first set up a model that describes the relation between a given particle size distribution and the resulting chord length distribution and then try to invert this model. The difficulty is that the inversion turns out to be an ill-posed problem. The idea of this work is to use not only information about the measurement procedure, but also about the crystallisation process. Process knowledge and measurement knowledge are combined by a Kalman filter with a constrained update step. By this approach, a direct inversion of the measurement equation is no longer necessary. (C) 2011 Elsevier Ltd. All rights reserved.