It really depends on what the line is about, and what you are trying to prove or negate.
Usually, if you have a lot of data, the line of best fits will be the line where the sum of squares of abberrants is the least. Abberant meaning the distance from any proposed line.
Any statistical program, even Excel, can give you this line.
If you have just a few points (<10) on your graph, it's different, because the influence of e.g. a faulty reading can become excessive. Normally you discard the lowest and the highest reading.
Did you like this example?