Although synonymous sites have historically been considered neutral, and are still used as a neutral reference for many analyses, several processes are known to exert a selective pressure on synonymous sites. Despite evidence for selection on synonymous sites across species and for several different processes, the extent of selection on synonymous sites and the relative contributions of the selective processes are not well understood. Using genome sequence data from two Drosophila melanogaster populations, we perform a SFS-based maximum likelihood estimation of purifying selection on fourfold degenerate synonymous sites using short introns as a neutral control. We estimate that ~15% of fourfold sites are under strong purifying selection. We find that the selection can largely be explained by codon bias and splicing-related factors. If preferred codons and alternatively spliced genes are excluded from the analysis, we find no evidence for selection, and recover the short intron level of polymorphism. These results clearly identify the major processes contributing to purifying selection on synonymous sites, and have implications for creating a neutral reference for other species.