This paper deals with the modelling and the simulation of an agglomeration in suspension process. It is shown how classical kernel models (constant, sum or product kernel) based only on the agglomerate size are inappropriate to represent such a complex process. A methodology for the elaboration of a phenomenological multidimensional kernel considering both liquid binding composition and size agglomerate is proposed. This kind of kernel is far more realistic than usual ones. The model parameters may be identified from experimental study. Preliminary results are presented and discussed.