http://www.gsmarena.com/gsmarena_lab_tests-review-751p6.php

Battery life
In gauging battery life, it's hard to perfect a methodology that lets you cover all of the various use scenarios without having to test for weeks on end. Whether it be display, audio quality, or camera, the goal with all of our tests is not only to provide you with the clearest, most comprehensive examination of a particular device, but to also provide it to you in a timely manner. Our battery test is no exception.

With each battery test you'll see four different numbers, each reflecting different types of battery-consuming activities. The test revolves around three basic tasks, which we feel are most commonly performed by smartphone users: making calls, browsing the web and playing video.

The first test is the talk time test, which measures how long it takes to deplete the battery by making voice calls. Bearing in mind that most screens automatically turn off during a call, we've made sure our set up accounts for this. We close all applications which may further strain the battery, too.

The web browsing test is performed using an automated script which reloads a webpage every ten seconds. There are no flash elements on the web pages, so the playing field is even. The brightness of the phones' displays is set to 50% and we use a Wireless N router placed a few meters away to get full connectivity bars.

In our video playback test we measure how long it takes for a device to run its fully charged battery down to 10%, while looping a standard-definition video. We stop at 10% since most devices shut their video players at this point. The display is again set to 50%, and all radios on the device are switched off (Airplane mode).

Finally, our overall longevity rating gives you an idea of how much battery backup you can get on a single charge. A score of 40h means that you'll need to fully charge once every 40 hours if you do one hour of 3G calls, one hour of video playback and one hour of web browsing daily. Essentially, it's an aggregated and normalized score, which factors in the three mentioned tests along with the real-life standby power consumption, which we also measure.