The goal of the quantification step is to score regions in the genome based on the number of reads. Recall that each transcript is fragmented into many smaller reads. Therefore, it is insufficient to simply count the number of reads per region, as this value would be influenced by (1) expression rates and (2) length of transcript. The higher the expression rate of a transcript the more reads we will have for it. Similarly, the longer a transcript is, the more reads we will have. This issue can be solved by normalizing the number of reads by the length of the transcript and the total number of reads in the experiment. This provides the RPKM value, or reads per kilobase of exonic sequence per million mapped reads.
This method is robust for genes with only one isoform. However, there is the possibility of overlap between conflicting variants of a transcript. When multiple transcript variants are involved, this problem is known as differential expression analysis. There are a few different methods for handling this complexity. The exon intersection model scores only the constituent exons. The exon union model simply scores based on a merged transcript, but can easily be biased based on the relative ratios of each isoform. A more thorough model is the transcript expression model, which assigns unique reads to different isoforms.