Mechanistic force prediction models require a calibration phase to determine the cutting coefficients describing the tool–target material interaction. The model prediction performance depends on the experimental correctness and representativeness of input data, especially in micromilling, where facing process uncertainties is a big challenge. The present paper focuses on input data correctness introducing a clear and repeatable calibration experimental procedure based on accurate force data acquisitions. Input data representativeness has been directly connected to the calibration window choice, i.e., the selection of the space of process parameters combinations used to calibrate the model. Also, the model validation has to be carefully carried out to make the model significant: the present paper proposes a clear and repeatable validation procedure based on the model performance index calculation over the whole process operating window, i.e., the space of parameters where the process works correctly. An objective indication of the model suitability can be obtained by applying this procedure. Comparisons among prediction performances produced by different calibration windows are allowed. This paper demonstrates how the calibration window selection determines the model prediction performance, which seems to improve if calibration is carried out where forces assume high values. Some important considerations on the process parameters role on cutting forces and on the model capability have also been drawn from the model validation results.