This thread seems to have taken on a life of its own. Since I have over 40 years experience as a professional astronomer (and now am retired), I feel my qualifications might be of some value to the discussion.
FIRST POINT: The entire algorithm for calculating sunset/sunrise times, in excruciating detail, can be found on Wikipedia under "Sunrise Equation". This, or another not significantly different, is the procedure the Naval Observatory uses. It is totally standard stuff. It does assume that the observer is at sea level, that his horizon is exactly 90° from the zenith (in other words, that his horizon lies in a precisely horizontal plane), and that standard meteorological conditions prevail. Any deviation from these ideal conditions will introduce a modest error in the result. Such error will increase with increasing distance from the equator, and can become quite significant at latitudes above, say, 50°.
SECOND POINT: Contributions to this thread suggest that Garmin units of different ages give age-dependent differences in SR/SS times. In order to see for myself what the effect might be, I rounded up 9 different Garmin hand-held units, differing in age over a range in excess of 10 years. Specifically, these were the following, with their approximate years of manufacture: (1) eMap (1999), (2) GPS III Plus (1999), (3) GPSMAP 76S (2002), (4) Geko 301 (2005), (5) Etrex Legend HCx (2008), (6) Etrex Vista (2009), (7) GPSMAP 60CSx (2006), (8) Foretrex 401 (2010), and Oregon 550t (2010). Today, I ran each one of these at my residence, and noted what the SR/SS times were from each unit for today (3 May 2011). Every last one (NO EXCEPTIONS!) gave sunrise at 06:13 MDT and sunset at 19:52 MDT. These can be compared with the results computed from the standard algorithm (FIRST POINT, above): sunrise at 06:13:14 MDT, sunset at 19:53:18 MDT. I see no meaningful difference from the result given by all the Garmin units.
THIRD POINT: In my earlier post, I suggested that Garmin might be omitting allowance for atmospheric refraction. I no longer see any justification for this suggestion, and I hereby withdraw it.
FOURTH POINT: I don't see any reason to stomp on Garmin's SR/SS calculations. I don't know how to account for the discrepancies reported by others.
End of diatribe!