Eugène [sorry, wrong name before],
one element of exposure control is understanding the latitude of the medium: how many stops under before the image disappears into "film base plus fog" or noise, how many stops over before highlights are "blocked" or "blown out".
Phil Askey at DPReview has made his third attempt at measuring dynamic range, which seems far better than his previous ones at least, using a Stouffer Step Wedge. Results are seen for the first time in his review of the FujiFilm S3 Pro, at
http://www.dpreview.com/reviews/fujifilms3pro/page18.aspOne thing I notice that fits with conventional digital wisdom: for both the S3 in normal mode (not using the special "R" pixels) and the 20D, the "latitude" is unbalanced, with about five stops below the mid-tones, only three above them. You probably know film behavior better than me, but -5 sounds better than film (good underexposure latitude), while +3 seems about as tight as reversal film, worse than negative film. (It also fits almost perfectly with the ISO definition of a sensor's base speed, which is based on placing the highlight blowout point at 170% diffuse reflectivity, or 3.2 stops above 18% gray.)
My reading would be that we should expose to control the highlights, much as with slide film, but with the reassurance that there is more latitude for underexposure when using low ISO settings, so it is usually not critical to get the highlights as far to the right as they can go.
That suggests a slight variation on Michael Reichmann's advice: expose to the right, but never go too far to the right, while being one or two stops to the left is usually safe.
P. S. Norm Koren has produced software called
Imatest which tries to measure dynamic range, amongst other things, but I have not studied it yet. I trust Norm better than Phil on technical issues though!
P. P. S. The S3 Pro's "R" pixels balance the range out by adding two stops of highlight headroom, which is just as Fuji claims, so I wonder what all the complaints are about!