【Objective】Variable rate fertilizing is an important way to improve fertilization economic efficiency and reduce adverse environmental effect. The objective of this paper is to make a diagnosis of the fertilization of wheat in Henan in an attempt to find out problems and to optimize application rates of NPK fertilizers in line with geographical change of wheat potential yield and soil properties.. Henan Province is a major wheat cultivation area of China, contributing 25% of the total wheat production of the country.【Method】Funded by the formula fertilization project of the Ministry of Agriculture of China, the province launched extensively Three-Zone Comparison Experiments (TCE) during the years of 2007—2009. TCE had experimental sites set up all over the province. Each site was designed to have three adjoining plots, randomly selected and laid out for wheat cultivation under CK (no fertilization), CF (local farmer customary fertilization) and RF (fertilization recommended by local agricultural experts), separately. In this paper, data were collected from a total of 2277 TCE sites in 82 counties for comparison analysis. Before seeding of wheat, a soil sample was collected from the cultivated-layer of each plot, and then the samples from the three adjoining plots were mixed into one as soil sample of the site for analysis of OM, pH, total N, Olsen-P and readily available K. Wheat yield, NPK application rates of each plot and soil properties and accumulated sunshine hour, accumulated precipitation, accumulated evaporation, accumulated temperature above 0°C and that above 10 °C of each TCE site, during the wheat growth period were all recorded separately for comprehensive analysis following the steps below. 1) mapping wheat yield, N, P, K application rate, contents of total N, available P and readily available K in soil with the aid of kriging interpolation, and comparing the maps to identify problems in application of N, P and K fertilizers in this area; 2) adopting the Random Forest(RF)method to assess relative contribution of soil properties of the cultivated soil layer, fertilization rate and geographical variation of soil type and climate to variation of wheat yield; and 3) superimposing the wheat potential yield map, soil total N map, soil available P map and soil readily available K map acquired in the research and referring to the research on soil nutrient sufficiency standard and recommendation of NPK application rates for targeted wheat yield to optimize fertilization recommendation geographically in line with the principles of “yield-targeted fertilization” and “soil-oriented fertilization” for the province. 【Result】Results show that the customary application rates of N and P tended generally to be higher, while that of K to be lower in some areas. However, most of the fertilization recommendations failed to reduce the rates of N and P. but generally raise the rate of K application. The two fertilization schemes, customary and recommended, are found to share a common problem, i.e. N, P, and K application rates mismatched targeted wheat yield and soil fertility in spatial distribution. Random forest analysis shows that geographic variations of soil, type and climate were the dominant factors of the spatial variation of wheat yield, and fertilization followed the next and soil nutrients the last in effect, while variations of geographical factors and fertilization were the ones determining spatial variation of wheat yield increment, which had nothing to do with soil nutrients. High fertilization rates were the major reason explaining the weak relationship between soil nutrient status and wheat yield.【Conclusion】On the background of high nutrient and water supply, physical geographical distributions of soil type and climate now are the key factors determining spatial variation of wheat yield at a provincial scale. Owing to the hardness of breaking through the obstacle of soil type and the limitation of climate conditions, it is almost infeasible to alter the rule of spatial distribution significantly within next few years. Based on the spatial variations of potential wheat yield and soil nutrient level, the recommended N, P₂O₅ and K₂O application rate should be in the range of 120~210 kg·hm^-2, 45~105 kg·hm^-2 and 45~120 kg·hm^-2, respectively. It is estimated that geographic optimization may save N, P and K fertilizer by 14.2%, 40.0% and 39.5% over the fertilization schemes recommended by local experts and by 20.9%, 41.1% and 17.5% over the customary application rate of the local farmers.