Dnmt3a-C, the catalytic domain of the Dnmt3a DNA-(cytosine-C5)-methyltransferase, is active in an isolated form but, like the full-length Dnmt3a, shows only weak DNA methylation activity. To improve this activity by directed evolution, we set up a selection system in which Dnmt3a-C methylated its own expression plasmid in E. coli, and protected it from cleavage by methylation-sensitive restriction enzymes. However, despite screening about 400 clones that were selected in three rounds from a random mutagenesis library of 60 000 clones, we were not able to isolate a variant with improved activity, most likely because of a background of uncleaved plasmids and plasmids that had lost the restriction sites. To improve the catalytic activity of Dnmt3a-C by optimization of the sequence of the DNA substrate, we analyzed its flanking-sequence preference in detail by bisulfite DNA-methylation analysis and sequencing of individual clones. Based on the enrichment and depletion of certain bases in the positions flanking >1300 methylated CpG sites, we were able to define a sequence-preference profile for Dnmt3a-C from the -6 to the +6 position of the flanking sequence. This revealed preferences for T over a purine at position -2, A over G at -1, a pyrimidine at +1, and A and T over G at +3. We designed one "good" substrate optimized for methylation and one "bad" substrate designed not to be efficiently methylated, and showed that the optimized substrate is methylated >20 times more rapidly at its central CpG site. The optimized Dnmt3a-C substrate can be applied in enzymatic high-throughput assays with Dnmt3a-C (e.g., for inhibitor screening), because the increased activity provides an improved dynamic range and better signal/noise ratio.