By checking the predictions of various types from the mean generating function time serial model,it is discovered that the mean generating function model is able to give satisfactory results,not only to the statistical values of observations (e.g. yearly or monthly means),but also to the instant values of observations (e.g. max temperatures),not only to the smoothly varied variables (e.g. temperature),but also to the violently changed variables (e.g.sunspot). Some methods to improve the mean generating function model are presented also.