The authors develop a predictive design method for synthetic ribosome binding sites (RBSs) to control protein expression levels in *Escherichia coli*. The method uses a thermodynamic model to predict the translation initiation rate based on the sequence of the mRNA transcript and the RBS. Over 100 predictions are validated, showing an accuracy within a factor of 2.3 over a range of 100,000-fold. The model correctly predicts that reusing an RBS in different genetic contexts can result in different protein expression levels. The method is demonstrated by optimizing the expression level of a protein to connect a genetic sensor to a synthetic circuit. The forward engineering approach will accelerate the construction and systematic optimization of large genetic systems.The authors develop a predictive design method for synthetic ribosome binding sites (RBSs) to control protein expression levels in *Escherichia coli*. The method uses a thermodynamic model to predict the translation initiation rate based on the sequence of the mRNA transcript and the RBS. Over 100 predictions are validated, showing an accuracy within a factor of 2.3 over a range of 100,000-fold. The model correctly predicts that reusing an RBS in different genetic contexts can result in different protein expression levels. The method is demonstrated by optimizing the expression level of a protein to connect a genetic sensor to a synthetic circuit. The forward engineering approach will accelerate the construction and systematic optimization of large genetic systems.