Many different kinds of data are available for modeling the specificity of a DNA-binding protein, and the quality of the model depends on both the type of data used and the algorithms for estimating binding energies. We discuss our approaches for modeling from several different types of data, and assess the accuracy of each based on experimental measurements. Given specificities for many proteins of a specific class one can also predict the binding specificities of novel proteins, allowing for the design of new proteins with unique specificities. We describe our current approaches to this challenging problem.