In sequencing results, the quality score is reported for each base, representing the probability that the base is called incorrectly. The notion of quality scores was initially developed for conventional Sanger sequencing, but is widely used for next-generation sequencing techniques, including Illumina. In this paper, we carry out in-depth analysis of quality scores reported for Illumina reads and present how they are related to real errors in the reads. We confirmed strong interrelation between quality scores and real errors in Illumina reads, and observed that reverse reads tend to have lower quality scores than forward reads in paired-end reads do. In addition, we discovered other interesting patterns from quality score analysis. Our hope is that the findings in this paper will be helpful for designing error-correction and/or filtering methods for next-generation sequencing.