Sensitivity is usually a function of range / resolution (plus all sorts of data on error, etc.). I'm not sure which 2g device you're asking about, but since you also posted in Triple Axis Accelerometer Breakout - ADXL345, just as a very basic comparison between the two:

The ADXL193 single axis accelerometer has a range of +-250g. Wikipedia on g-force suggests that this means you could measure 'Brief human exposure survived in crash'. It is an analog output device, so it doesn't necessarily have an inherent 'resolution'. However, for most purposes you'll have to convert this to a digital value through an ADC somehow. If you're using an Arduino, then the ADC is 10 bits, meaning you get about 1024 levels over this range. Unless you do some analog tricks to modify the linear relationship between g force and output voltage, this means you get a resolution of roughly 0.5g/level.

The ADXL345 triple axis accelerometer has a range of +-16g. Wikipiedia on g-force suggest that this means you could measure 'Typical max. turn in an aerobatic plane or fighter jet'. It has a digital output with a 13bit resolution. Here you don't have to use an ADC, but will of course have to read the data correctly. The resolution here would be roughly 0.004g/level.

So if you're asking which one is more sensitive, the ADXL345 clearly wins. It's also triple axis, and a few bucks cheaper. The ADXL345 is an option if you need to measure accelerations outside of the ADXL345's range (such as its use suggested in the datasheet: for car impact airbag purposes).

For some further light reading, see also:
How to Select an Accelerometer
Accelerometer, Gyro and IMU Buying Guide

Either, really. The arrow is defining the axis. As long as the force is occurring in parallel with the arrow, you're going to get a reading. Make sense?

Hi cobbr2,

I'm trying the same with a diabolo. Did you have any succes with the yoyo? I assume you wanted to make a POV led display in the yoyo? Ho did you manage to calculate the direction of maximum acc with only one sensor? I think its impossible and you need at least two opposoed to eachother and substract their readings to cancel out the centrifugal acceleration

I wouldn't call it "the wrong way" The chip appears to output a nominal 2.5v with no force acting on it. A positive acceleration will bring it to 5v while a negative acceleration will bring it to 0v. Either way you'll get the same change as far as the readings go, you just might have to flip the sign somewhere in your code.

Although flipping the silkscreen arrow would be "more technically correct". :P

According to the datasheet, when the supply voltage is 5V, the part will output about 8mV (0.008 V) per g. Zero g will be approximately 1/2 of the supply voltage, but could vary so you should test the individual part at rest and use that voltage as zero g. So if the supply is 5V, zero g will be around 2.5V, and 100 g will be around 3.3V (2.5V + (0.008V * 100g)). The output is relative to the supply voltage, so if you're running it at a different voltage you'll get different results.

Like WizenedEE says, your measurement error will be huge. That's because to go from acceleration to distance you have to integrate twice, and you'll integrate the error as well.

Besides, due to this sensor's low sensitivity slow acceleration will hardly be detected, resulting in an even greater error.

You'd need an accelerometer with astronomical precision to accomplish that. It is technically possible, but there would be a ton of error.

Not really, since you'd have to have your uC constantly checking it for data. You could have it wake up every second or so and check, then go back to sleep, but then you'd have to make sure your large acceleration happens for more than 1 second.