Probelm in HTSeq counts using pair-end reads

[wenchen]

Dear all:
I have a question regarding HTSeq-count. Not sure if this is the right group to ask, but I thought I would give a try first here.

I have been using HTSeq-0.6.1p1 (an old version) to calculate all the counts after Tophat mapping using Illumina NGS platform. When my reads are single-ended, everything works great!
Now for the 1st time my reads are pair-ended, the issue starts.

Frist, I got all the warning signs saying "XXXXX reads with missing mate encountered". But I still got the counting results at the end, which to be honest looked fine to me.
Then I found out quickly that I need to sort the bam files. So I did:

samtools sort -n accepted_hits.bam accepted_hits_sorted.bam
samtools view -h accepted_hits_sorted.bam accepted_hits_sorted.sam
htseq-count -t CDS -m intersection-strict -r name accepted_hits_sorted.sam Homo_sapiens.GRCh37.75.gtf >HTSEq.output

Presumably the new results after sorting should be the same, or at least similar to the one without sorting. But I was very surprised to find out the the new counting result after sorting is only about half or less that the result without sorting.
Has anyboday seen this before? How to explain this result?

Thanks very much for your help!
Joan