Tóth,T., S. Matsumoto, R. Mao and Y. Yin. 1994. Plant Cover as Predictor Variable of Salinity and Alkalinity in Abandoned Saline Soils of the Huang-Huai-Hai Plain, China. Agrokémia és Talajtan. 43:175-195. The correlation between the seminatural vegetation cover and soil properties is useful for predicting soil properties in the abandoned saline soils of the Huang-Huai-Hai Plain of China. For the mapping of soil properties we propose a basic study of the correlation and spatial distribution of the important soil properties and plant cover. Based on it, it is possible to select variables of plant cover and other easily available field measured properties as predicting variables of soil properties. These predicting variables can be used in regression equations and cokriging to improve the prediction of soil properties. In an area of 100x220m, large blocks (n=55) of 20x20m size and medium blocks of 5x5m size have been used for prediction of soil pH and soil salt content. The correlations existing between soil properties and easily measurable plant cover, surface elevation and penetration resistance showed that the main plants of the abandoned land, Phragmites communis and Imperata cylindrica have distinctive preference for the soil properties. The best predicting variables of soil properties were artificial variables, factor scores derived from the measured plant cover, the cover percentage of the two most important plants, Phragmites c. and Imperata c. and soil penetration resistance. The two block sizes showed comparable precision in the estimation of the soil properties, and the size of the large block was reasonable in the mapping of the fertility of the abandoned plots. Kriging was more precise in the larger, averaged blocks and regression was more precise in the medium blocks. Kriging and cokriging with plant factor scores, plant cover and penetration is more precise than regression analysis for the estimation of soil pH and salinity and should be used for estimating the soil properties always when there is large enough uniform plot to do it. When the blocks to be mapped are small and not contiguous, multiple regression with easily measurable properties can be used with similar precision.