In a previous directed evolution study, the B-FIT approach to increasing the thermal robustness of proteins was introduced and applied to the lipase from Bacillus subtilis. It is based on the general concept of iterative saturation mutagenesis (ISM), according to which sites in an enzyme are subjected to saturation mutagenesis, the best hit of a given library is then used as a template for randomization at other sites, and the process is continued until the desired catalyst improvement has been achieved. The appropriate choice of the ISM sites is crucial; in the B-FIT method the criterion is residues characterized by highest B factors available from X-ray crystallography data. In the present study, B-FIT was employed in order to increase the thermal robustness of the epoxide hydrolase from Aspergillus niger. Several rounds of ISM resulted in the best variant showing a 21 °C increase in the T(60)(50) value, an 80-fold improvement in half-life at 60 °C, and a 44 kcal mol(-1) improvement in inactivation energy. Seven other variants were also evolved with moderate yet significant improvements; these were characterized by 10-14 °C increases in T(60)(50), 20-30-fold improvement in half-lives at 60 °C and 15-20 kcal mol(-1) elevations in activation energy. Unexpectedly, in the ISM process the best variants were obtained from essentially neutral or even inferior mutant parents, that is, when a given library contains no improved mutants. This constitutes a practical way to escape from what appear to be local minima ("dead ends") in the fitness landscape-a finding of notable significance in directed evolution.