The stop codon readthrough during translation has been widely reported and is in general deleterious unless in virus which has a compacted genome and uses it as a potential mechanism to diversify proteome. Here, taking advantage of high-throughput sequencing technology, we systematically studied the impact of sequence context on the level of readthrough. To this end, we constructed a plasmid with the expression of a fusion protein of dTomato-GFP from TDH3 promoter in Saccharomyces cerevisiae. A stop codon was inserted between dTomato and GFP, together with 9 bp random sequences at both sides. We dissected the readthrough into two aspects, readthrough probability and readthrough strength, which represent the proportion of cells with readthrough in the population and the average degree of readthrough in the cells with readthrough, respectively. We found that the variance of readthrough probability is much larger than that of readthrough strength, suggesting that natural selection may mainly operate on readthrough probability. Interestingly, the adenosine following stop codon leads to low readthrough probability and high readthrough strength, which causes abundant readthrough in a small fraction of cells, enlarging the cell-to-cell variation on stop codon readthrough in a cell population. We further analyzed the first nucleotide in the 3’UTR of yeast genome and found that adenosine is the most frequently used nucleotide, suggesting natural selection on translational readthrough constrains the evolution of 3’ UTR. Taken together, our study distinguished two independent aspects of translational readthrough, identified the coding rules of translational readthrough in the genome, and shed light on the genomic evolution of 3’UTR in yeast.