A common practice is to visually inspect the data, and remove outliers by hand. The problem with this approach is that it is arbitrary. It is too easy to keep points that help the data reach the conclusion you want, and to remove points that prevent the data from reaching the conclusion you want.

The graph above was created via simulation. The values in all ten data sets are randomly sampled from a Gaussian distribution with a mean of 50 and a SD of 15. But most people would conclude that the lowest value in data set A is an outlier. Maybe also the high value in data set J. Most people are unable to appreciate random variation, and tend to find 'outliers' too often.