There's a misconception among some runners that their GPS watches are above reproach. You sometimes hear of runners complaining that a course was marked too long because their own Garmins ran a few tenths of a second longer than the posted race length. It'll even crop up in race reports on popular forums, on reviews on Yelp of MarathonGuide.com or on blog posts.
Almost always, the discrepancy between the GPS reading and the course length is due to the fact that certified courses are measured at the shortest possible distance with respect to corners and tangents, and because no one can reasonably run that absolutely shortest possible distance unless they are fast enough to go wire-to-wire in first place and have a brilliant mind able to see angles and tangents needed to take. Runblogger D.C. Rainmaker has gone into fabulous detail on this issue in the past.
But the New York Times last week took a different angle on why you shouldn't trust your Garmin: because Google Maps spits out a different number.
The problem, of course, is that web services like Google Maps are largely constructed as navigation tools, not accurate-to-the-tenth distance plotters. Running watches are designed as distance plotters first and foremost. As Rainmaker notes in a post on the Times' story, Google Maps in particular is measuring routes from the middle of a road, while most runners are using the shoulder or sidewalk. No wonder there'd be a discrepancy when you compare the Google Map route length and the GPS readout. You're running different routes according to each.
The Times story also brings up GPS' issues with trail runs; I would not encourage runners to rely on Google Maps to plot out an accurate-to-the-tenth trail run. Even using a made-for-runners offshoot like G-Map Pedometer gets a little aggravating after, you know, two or three switchbacks.
GPS watches do indeed have trouble at the track, but chances are that you're using the watch more for its timing capability than its distance accuracy when you're doing a workout on a short, measured course. (Is anyone at the track basing their quarter-mile repeats on when the watch spits out 0.25 miles instead of when they return to their starting point?) As Rainmaker notes, services like Google Maps are indeed great for plotting out routes in advance, to make sure you're not setting out on an eight-mile loop when you mean to run six. But measuring out a six-mile loop on Google Maps and running a basic stopwatch isn't going to give you any more an accurate distance or pace than a Garmin would. There are reasons to believe your Garmin is off up to 2 percent on distance, but there is no reason to believe a web map is any more accurate.