A general approach is described for creating a comprehensive and quantitative model for predicting the binding of a monoclonal antibody (mAb) to any short linear sequence of amino acids. Binding of 8 different mAbs with continuous epitopes to a common library of ~122,000 near-random peptide sequences arrayed on a silica substrate was used to train neural network models resulting in a general sequence-binding relationship for each mAb. The model was then used to predict mAb binding to 1,000,000 randomly generated sequences as well as the cognate sequence of each mAb. In every case the cognate sequence ranked in the top 250 sequences, and for 5 of the 8 mAbs it ranked in the top 10 out of one million. Practically, this approach has potential utility in selecting highly specific mAbs for therapeutics or diagnostics. Conceptually, this demonstrates that very sparse, but near random, sampling of a large sequence space is sufficient to generate comprehensive models predictive of highly specific molecular recognition.