A reduced model of a urea crystallizer is developed for process control purposes. It is derived from a reference model, which describes the formation of particles in fluid flow and is of very high order. A strong reduction of the system order is achieved by proper orthogonal decomposition (POD). However, it turns out that POD alone does not lead to a satisfactory reduction of the computation time. The reason is the presence of nonlinear terms in the reference model whose evaluation is quite costly in the reduced model. To get a speed-up of the reduced model, the nonlinear terms are approximated by an interpolation technique. The resulting reduced model is 500 times smaller and 100 times faster than the reference model, while the loss of accuracy is marginal.