Abstract: The motivation of the study is to discuss the issue whether numerical weather prediction should be practiced in the meteorological departments at provincial, regional or city levels. It suggests that numerical weather prediction need not be operated by meteorological departments at all levels, and only need to be done by those who carry on scientific-research and urgently expecting to improve the accuracy of local high-impact weather prediction. With respect to the problem how numerical weather prediction should be put into practice, it is put forward that local meteorological departments are not fit for repeating those approaches originated from mainstream opinions, but should break a new path such as an alternative methodology.Furthermore, the content, method and meaning of such an alternative methodology are examined. The inverse problem that the information of historical data including analogue is made full use of estimating model errors is raised and an analogue errors correction method is theoretically introduced in order that one can combine effectively both statistical and dynamical methods together, and need not change the current numerical prediction models. The new alternative methodology not only adequately utilizes dynamical achievements but also can rationally extract the information of a great many analogues in historical data so as to reduce model errors and improve forecast skills. This prediction strategy can overcome the limitation of only using initial values in traditional numerical models on one hand, and on the other hand, is able to realize the pertinence of numerical model prediction for specific elements through the process of selecting their corresponding analogues.Theoretical analyses show that based on analogue correction method of errors, the conventional problem of numerical weather prediction may be transformed into the prediction problem of forecast errors. That is to say, the unknown error terms in current prediction can be mathematically estimated by means of the forecast error fields generated from historical analogue data similar to the current prediction initial values. Some related preliminary experiment results on a complicated atmospheric prediction model have documented the feasibility and validity of this method based on such alternative methodology.As are seen in related studies, in such an alternative methodology as analogue correction method of errors, it emphasizes that it is unnecessary to construct new numerical prediction model which is very difficult for local meteorological departments. In such circumstance, one only needs to operate the existing models and modify them in order to localize by using historical data associated with interesting phenomena. There for it is completely feasible in practice and has operational future. Of course, so far the selection process of historical analogues is still coarse, and especially, as to the problem how the unknown error terms in current prediction can be estimated using the forecast errors from historical analogue data requires much more work, and further researches are still needed.