Recommended prep. technique of curves?

[miro2009]

What is the most recommended technique for preparing a calibration curve for a RP-HPLC method, serial dilutions or preparing each point seperatly? I'm trying to compare both methods using the same set of pipettes (calibrated) by comparing the peak areas of each equivalent prepared levels, and observe that at higher concentrations, there are very slight differences, but as the concentrations get lower the differences become very significant and accordingly I obtain 2 curves with different slopes and still both are linear with R2 0.999. I could try to evaluate the curve by a seperate preparation of the standard in the range of my target (middle of the curve, but still how should I prepare it, with serial or a single dilution (1:20)?
Are there any special settings also that should be done when constructing the curve in addition to linear best fit? for eg. I don't use "force to go through zero" option, is that correct?
Also, is it acceptable to have a curve of 9 points or is that too much?
Awaiting your opinions
If anyone has links to articles, books discussing these fine details I would appreciate the help..

[tom jupille]

Serial dilution is not recommended (by me, anyway!) because errors accumulate (especially bias errors).

Never force a linear regression through zero.

Nine calibration levels is certainly acceptable, but is probably overkill. Five is sufficient for most purposes.

The best tutorial on statistics that I know of is Coleman & Vanetta's series in American Laboratory. The articles are available on-line from ISC's web site:
http://tinyurl.com/yzewng3