Well...you can't, not really.

RNA integrity and efficiency of cDNA synthesis can have large effects on measured expression values: the whole point of normalising data to a set of reference genes is to account for this.

Without such normalisation, you cannot distinguish a difference in expression from a difference in cDNA synthesis efficiency, for example.

The only cases where I would (hesitantly) accept that ref genes are not an obligate requirement would be under scenarios where you already have ample evidence that RNA isolation and cDNA synthesis is highly reliable in your hands, and you have a LOT of samples, and are looking at expression differences of one or more orders of magnitude.

As a reviewer, however, I would still require a reference gene (ideally two or three).

Ok , I got it

But I didn't normalize to a calibrator .

I already have house keeping gene.

If you used a reference gene like GAPDH, then all of your RT data from your target genes should be 'normalized' to the data from the reference gene - otherwise it isn't really RT PCR.

If you are trying to compare one experiment to another experiment and your raw numbers are very different, convert your data to fold change, and then you can compare experiments to each other.