The A/D converters have a FET switch on each input. When you change the channel, the gates on those FETs toggle. There is a finite capacitance between the gate and the drain and source of the FET, so the toggling of the gate signal causes what is called charge injection into the other pins. This causes a glitch at the input and output of the FET switch. If you read the A/D too quickly after you change the channel, you will get an inaccurate reading due to voltage spike caused by the charge injection. The time it takes for this excess charge to bleed away is dependent on your source impedance. If your source is the output of an op-amp (zero ohm source), then you can switch channels and start conversion within 3uSec. If your source is 10K then you should wait at least 11uSec after changing channels before you start a conversion. If you don't wait long enough, your accuracy will suffer.