Modern management and machines have been introduced on a demonstration farm in Ganhe (China). This has led to new methods of cultivation with effects on yields, cost structure and thus also on the economic success of the farm. These effects should be tested with the help of an on-farm trial. The cultivation methods differed in the equipment used, plant protection and fertilisation strategies. In contrast to classical field trials, normal working practice farm machinery and fields are used in on-farm research. Thus it is not usually possible to avoid the trial fields showing a soil trend as well as spatial correlations, which influence the estimated values for expected value and variance of the cultivation method. Thus, through the use of modelling with the “procedure mixed” in the statistical software SAS 9.2, an effort is made to take into account these influences in the statistical evaluation of agricultural field trials. An accordingly important model extension is possible based on the theory of the mixed linear model. The goal is to take into account the spatial trend and correlations related to the data and thus to be able to guarantee that statistical risks are kept to. In order to record the heterogeneity of the soil a measurement took place before the beginning of the trial to determine the apparent electrical conductivity of the soil using the soil scanner EM 38. The paper shows how soil trends and spatial correlations can be dealt with and how the evaluation results can vary depending on these. A high heterogeneity of the trial field is found particularly in precision farming trials, since these are also a part of the procedures to be tested. It is shown that, combined with suitable statistical evaluation procedures, on-farm research is an instrument for carrying out precision farming trials.