Codon usage has been reported to be correlated with local recombination rate, which can be explained by two proposed models. In the present study, correspondence analysis was used to investigate the major trends in codon usage variation among S. cerevisiae genes. It was found that the first principle source of codon usage variation in yeast is due to the variance of expressional levels, which is consistent with the previous translational selection model. Moreover, recombination rate is also correlated with the codon pattern, which might be a byproduct of mutational bias associated with recombination rather than the consequence of Hill-Robertson interference. A recent study has analysed the genome sequence, but reached opposite conclusions: the positive correlation between recombination rate and codon bias in yeast mainly results from Hill-Robertson interference. In light of this conflicting result, we have discussed the possible reason and found that the previous analysis was undermined by mistaken assumptions that weak selection acting at expression level led to the correlation between recombination and codon bias.