Abstract: Dimensional analysis and reduction are made to the two commonly used schemes of 4th order linear horizontal diffusion, as well as the detailed control experiments between the two schemes are made using a tepography-included mesoscale model. Horizontal diffusion is calculated on σ-surfaces in one scheme (known as Scheme A afterwards), and on p-surfacas in another (called Scheme B hereafter). Experiments show that, differences are small in smooth-terrain areas and very large in steep mountain areas where the 24h rainfall prediction deviation exceeds 50mm, between the forecasts of the two schemes. The reason for that is that temperature and humidity are falsely diffused in Scheme A, which causes abnormal temperature and humidity,and results in the abnormalies of.unstable layer and convective developement. In addition, Scheme A causes circulation anomalies, which is connecting directly to the convective abnormality, which in turn may affect the arrounding areas through turbulent wave, causing rainfall prediction deviations in the area. Analysis indicates that,Scheme B, which can minimize the diffusion scheme involved forecasting abnormal in steep mountain areas and arround, is necessary and feasible.