Nick Loman at Pathogens: Genes and Genomes has some tips for de novo bacterial genome assembly. He says that Velvet is "an excellent option for short-read de novo assembly of bacterial genomes" but "getting the best out of Velvet is a bit trickier than using Newbler." His four steps are to "count per-base quality scores and trim the reads," "eliminate singletons," "chuck away reads with Ns in," and "throw out reads with low average quality."