This paper analyses methodologies to calibrate and validate probabilistic lane change models. We perform a calibration and validation on lane change models (microscopic and macroscopic) which take the most basic dependencies into account. The resulting model has reasonable parameters, and the goodness of fit for the validation set (hold back from the total set) is similar to the calibration. For two measures of validation the model hence is validated. However, in real world terms, the model performs quite bad. It is hence concluded that the model should be validated based on measures which have a clear physical interpretation, and based on those the quality should be judged.