In this paper, generic bandwidth-scalable behavioral models suitable for power amplifiers exhibiting memory effects are proposed. The models are built around state of the art two-box models, namely the Hammerstein model and the forward twin-nonlinear two-box model, and take advantage of the separation, in these two-box models, between the static and dynamic distortions of the power amplifier. In the proposed bandwidth-scalable two-box models, rather than updating the entire model coefficients when the signal bandwidth changes, the memoryless function is maintained unchanged and only the function modeling the dynamic distortions is updated. Experimental validation carried out on a Doherty amplifier prototype shows that the proposed bandwidth-scalable models are able to achieve the same performance as the conventional models while reducing the number of parameters to be updated following a change in the operating signal bandwidth by up to approximately 75% for the Hammerstein model and 40% for the forward twin-nonlinear two-box model. The developed models are suitable for emerging wireless applications where operating conditions varies rapidly.