This paper blends the ideas of estimation of model-plant uncertainty from process data with the robust design of generalized predictive control (GPC) based on the small-gain theorem. Classical signal processing methods with appropriate smoothing techniques are used for the estimation of model-plant uncertainty and its upper bound for SISO linear time-invariant systems. The choice of proper input excitation in the estimation of uncertainty spectrum in the presence of additive noise is also discussed. The estimated uncertainty spectrum with the stability bounds are then used to provide simple graphical based tuning guidelines for robust design of GPC. A more accurate estimate of the uncertainty bound allows one to trade-off robustness with performance. The use of important GPC tuning parameters that allow one to shape the stability bound for robust performance is highlighted. Evaluation of this technique is illustrated via simulation and experimental application.