With the increasing availability of high resolution data, remote sensing is gaining importance for agricultural management. Sensor constellations such as RapidEye or Sentinel-2 have a strong potential for precision agriculture because they provide spectral information throughout the cropping season and at the subfield level. To explore this potential, methods are required that accurately transfer the spectral information into biophysical parameters which in turn permit quantitative assessments of plant growth on the field. Boundary condition for a successful monitoring, e.g., a repeated derivation of the biophysical parameters is to cope with the challenge of enormous data amounts, i.e. to select the input data that is most relevant.In this study, biophysical parameters of winter wheat, namely the fraction of absorbed photosynthetic active radiation (FPAR), the leaf area index (LAI) and the chlorophyll content (expressed by SPAD), were modelled with RapidEye data in Mecklenburg-West Pomerania, Germany, using Random Forest based on conditional inference trees. Focus was set at the selection of the most important information out of spectral bands and indices for parameter prediction on winter wheat. Insitu and remote sensing observations were grouped into phenological phases in order to examine the importance of single spectral bands or indices for modelling biophysical reality in the several growing stages of winter wheat. The coefficient of determination for FPAR (LAI; SPAD) ranged between 0.19 and 0.83 (0.33 and 0.66; 0.21 and 0.45). Model accuracy was linked with the phenological phase. The results showed that for each biophysical parameter, different spectral variables become important for modelling and the number of important variables depends on the phenological time span. The prediction of biophysical parameters for short phenological groups often depends only on one to three variables. The results also showed that in the phenological phase of fruit development, the model accuracy is the lowest and the determination of the importance is comparatively vague.