A modified version of the widely used CERES-Wheat phenology model (CWm) was developed introducing variable thermal time from double ridge stage to flag leaf emergence and compared with the original model (CW V3.x). Both model versions were newly implemented and calibrated using routinely collected field ratings of phenological stages of winter wheat crops. Calibration for all parameters of both model versions was done using a step by step procedure thereby using different methods for parameter identification. For calibration and validation of the model a data set containing more than 6000 single observations of wheat phenology was used. The improved model version had a better prediction of phenological (BBCH) stages compared to the original CW approach for an independent validation data set. The average RMSE for BBCH 37 and BBCH 39 was decreased by two and five days, respectively. We could show that it was possible to calibrate the wheat phenology models using solely routinely available field ratings. Because the predictive quality for wheat phenology of actual wheat crop growth simulators without calibration especially often is still limited the presented approach may be seen as an approach for improving this type of models without especially designed experiments.