Yes, you can align against known mRNAs, and, if you have a novel exon via alternate splicing, align against genomic DNA. This will work if your RNA-Seq data contains an allele of a SNP that is different from the reference genome - you'll see that difference and it may align with a known SNP.
On the other hand, the source of your RNA-Seq data may contain only the reference allele of the SNP, matching with the genomic or RefSeq mRNA perfectly. In this case, you'd like to align or compare to known SNPs to see what is present.
The third possibility is one where the source is heterozygous - two alleles at one position. Exactly how you detect this depends on how your RNA-Seq data were assembled into mRNAs. If the two alleles, say A and G, are each expressed at roughly the same levels, does the mRNA carry an A, G, R, or N at that position? What about the situation when you have allele-specific differences in expression of the gene, where the A level has 9 transcripts for every one with the G. Perhaps your assembler called an A, when in reality the situation is: G is present, was detected, but is less stable/preferred. If you are aligning the individual reads and see this allele difference, the assessment is much easier.
Sounds like you have figured it. However, since you are looking at RNAseq, I recommend you use a splice-aware tool like tophat/bowtie for alignment. This will allow your RNAseq reads which span exon junctions (gapped alignments) to also be viewed and should result in better total coverage. I don't think eland does a terribly good job aligning RNAseq data to the reference genome.
I was also looking for the mutations in RNA seq data. I used Tophat/bowtie and samtools for data. The .bam file include the variants (including both SNP's and mutants) How can we differentiate them?
Tophat gives great output files (.bed files of insertions, deletions and junctions) and it can be viewed in UCSC genome browser. Through these .bed files, is it possible to detect the mutations only?