P-splines were introduced by Eilers and Marx (1996). We consider semiparametric models where the smooth part of the model can be described by P-splines. A mixed model representation is also considered. We set a simple strategy for the choice of P-spline parameters, ndx, bdeg and pord, and discuss the use of various criteria for smoothing parameter selection. We illustrate our remarks with the analysis of a randomised block design.