Replicates
We strongly advise all investigators to include at least three biological replicates for each experimental condition, including control. By including replicates in your experiment, statistical approaches can be applied to your data analysis. There is always variation between individual research subjects, some are small, i.e. cell lines, whereas others are large, i.e. patient samples. One sample cannot be used to represent the entire population. Thus replicates are required.
To maximize the benefit of these extra samples, you should choose your replicates carefully. For instance with animals, they should be the same sex, preferably littermates. They should be raised in the same conditions and given the same treatments at the same time. They should be as genetically similar as you can make them. Obviously, it can be difficult in a large experiment to meet every single one of these conditions. Care should be taken so that your replicates are as similar as possible.
Replicates should be biological replicates. Technical replicates (i.e., hybridizing the same sample on three different arrays) will only assess variations in manufacturing, dye incorporation, pipetting, etc. These variations tend to be much smaller than the real variations between organisms, and they can be corrected through data normalization. Therefore, the most value is achieved through the proper use of biological replicates.