This paper treats the problem of synthesizing antiwindup compensators that are able to handle plant uncertainty in addition to controller saturation. The uncertainty considered is of the frequency-weighted additive type, often encountered in linear robust control theory, and representative of a wide variety of uncertainty encountered in practice. The main results show how existing linear matrix inequality based antiwindup synthesis algorithms can be modified to produce compensators that accommodate uncertainty better. Embedded within these results is the ever-present performance-robustness tradeoff. A remarkable feature is that the often criticized internal model control antiwindup solution emerges as an "optimally robust" solution. A simple example demonstrates the effectiveness of the modified algorithms.