This paper demonstrates the use of a geotechnical drum centrifuge in the calibration of a model to predict the peak punch-through penetration (qpeak) resistance of a spudcan on sand overlying clay. A series of loose sand overlying clay tests was performed and combined with an existing database of tests performed on dense sand overlying clay. The performance of the failure stress dependent model proposed by Lee (2009) and Lee et al. (2009) has then been assessed using this combined dataset which encompasses a wider range of soil properties and problem geometries than was used in the original calibration of the model. The single empirical stress distribution factor (DF) that is employed in the model was then optimised using a back calculation procedure for all tests. The scatter of the optimised DF values was then compared for the original bi-linear calibration proposed by Lee et al. (2009) and a new nonlinear power law calibration. The new non-linear relationship enables the model to better predict qpeak over a wider range of problem geometries and for conditions of both loose and dense sands overlying clay. The work demonstrates the importance of the geotechnical centrifuge in calibrating such models given that at present numerical methods are unable to reliably capture such punch-through behaviour and good quality field data of punch-through failure of spudcan foundations is unavailable