I have to validate a method for residual solvents analysis, with an analytical range of 60 (from 0.1% to 6% of solvent content).
I can choose between a "single-point" calibration (three standards at a single concentration level) or a three level calibration (three standards ranging within the quantitation range).
Which one should I choose? I don't feel "comfortable" with the single-point, but, actually, both methods are linear in the analytical range.
Can anybody tell me which is the most commonly accepted practice with this kind of analysis?

It is my understanding that if you have linearity and near zero intercept on the x-axis, a three point calibration is preferable and less likely to be contested by a reviewer.
I would use a point at or below the lowest level of measurement and one above the highest level of measurement. But a question, why only 3 points, why not 5 points. Two outside the measurement range and 3 points within sounds even better to me.

If you are doing a USP type of residual solvent analysis the most common method, 467, calls for a single point calibration. Without getting into the overall quality of the method, it is set up as a basic pass/fail situation. If all you are doing is pass/fail, then a single point calibation at the action level is all you need, assuming reasonable instrument stability. If you want to do more than a simple pass/fail test, then a multipoint calibration is better. 3 vs. 5 vs. more is an interesting statistical question, but for a less than 2 order of magnitude range 3 points will do fine, again assuming linear response over the calibration range.

1000 ppm to 6% is certainly not a USP test.
I believe the poster desires an accurate measurement of the solvents which may be necessary to determine std purity of a drug or a related, or to assist in process improvement.
If you want confidence in your determination and its application to whatever use you desire demonstrate not guess its actual content. That requires a high recovery below the lowest level of measurement and a demonstration of linearity through and past the high measurement point which should have 3 or more additional points. A 5 point regression as I posted above is performing a thorough and nearly uncontestable determination. Do the points curve? Do you recover 90% of an additional 1000ppm spike at 1000ppm level of contamination? Do it right the first time so no one can contend with you about your results. IMHO
:-) But you can always do it over. Job security and all that.

Thank you all for your answers.
I actually need to perform a quantitative analysis, not a simple limit test, that is the reason for my preference for a 3 point calibration. I wanted to be sure that somebody other share my opinion: now I feel more "comfortable"!
I agree that 5 standards would give narrower confidence limits on the results, but I have to make a compromise between the number of samples and that of standards.