In order to improve the quality of de novo assembly of next-generation sequencing data, we provide treatments for the data pre-processing stage by quality trim, error correction and random shuffle. All of the treated data are assembled by three tools: Velvet, SOAPdenovo and ABySS. A fractional factorial design is implemented to reduce the runs needed to find the proper parameters of the tools. We validate the treatment effects by the alignments results of bacteriophage Phix 174, whose genome is well studied. Our results confirmed that quality trim and error correction will provide essential improvements to de novo assembly. After quality trim, random shuffle of the reads may not lead to any improvements by using SOAPdenovo and ABySS. However, random shuffle did improve the results of using Velvet alone.