Bacillus subtilisin has been a popular model protein for engineering altered substrate specificity. Although some studies have succeeded in increasing the specificity of subtilisin, they also demonstrate that high specificity is difficult to achieve solely by engineering selective substrate binding. In this paper, we analyze the structure and transient state kinetic behavior of Sbt160, a subtilisin engineered to strongly prefer substrates with phenylalanine or tyrosine at the P4 position. As in previous studies, we measure improvements in substrate affinity and overall specificity. Structural analysis of an inactive version of Sbt160 in complex with its cognate substrate reveals improved interactions at the S4 subsite with a P4 tyrosine. Comparison of transient state kinetic behavior against an optimal sequence (DFKAM) and a similar, but suboptimal, sequence (DVRAF) reveals the kinetic and thermodynamic basis for increased specificity, as well as the limitations of this approach. While highly selective substrate binding is achieved in Sbt160, several factors cause sequence specificity to fall short of that observed with natural processing subtilisins. First, for substrate sequences which are nearly optimal, the acylation reaction becomes faster than substrate dissociation. As a result, the level of discrimination among these substrates diminishes due to the coupling between substrate binding and the first chemical step (acylation). Second, although Sbt160 has 24-fold higher substrate affinity for the optimal substrate DFKAM than for DVRAF, the increased substrate binding energy is not translated into improved transition state stabilization of the acylation reaction. Finally, as interactions at subsites become stronger, the rate-determining step in peptide hydrolysis changes from acylation to product release. Thus, the release of the product becomes sluggish and leads to a low k(cat) for the reaction. This also leads to strong product inhibition of substrate turnover as the reaction progresses. The structural and kinetic analysis reveals that differences in the binding modes at subsites for substrates, transition states, and products are subtle and difficult to manipulate via straightforward protein engineering. These findings suggest several new strategies for engineering highly sequence selective enzymes.