A new procedure has been developed to minimise the influence of the sample imperfections on the identified material parameters by eliminating from the objective function those modes that are mostly affected by the model errors so that eventually no modifications of the numerical model are required. The model errors in the present study are defined as deviations of a test sample from the assumptions made in the numerical model, namely, assumptions about lamination angle, plate flatness and uniform thickness. To improve an accuracy of the error functional coming from the physical tests, an influence of measurement errors on the modal parameters has been investigated additionally. The proposed practical solution makes the results of inverse problems more accurate and reliable and the identification process more effective. This new methodology has been successfully applied for the non-destructive material properties characterisation of carbon/epoxy prepreg (SEAL® Texipreg HS 160 RM) panels.