For semiparametric regression model selection, based on a model selection criterion there is no finite order (or number of parameters) of the nonparametric part to be estimated consistently, but there is a finite order (or number of predictor variables) of the linear part to be estimated consistently.