this site might be useful too :)
http://www.alcula.com/calculators/statistics/linear-regression/
If you want to get into the technicalities of this, start with wikipedia.
If you have a manageable number of data points plotted on graph paper, draw a couple of trial straight lines that look like they may be good fits. One at a time, for each line, measure and add up the "errors", i.e., the discrepancy between each plotted data mark and the corresponding point on the straight line fit. Finally, whichever straight line has the least total "errors" is the best fit!
In practice, mathematicians usually go for a more complicated way to evaluate the "errors" in the fit. (Well, they wouldn't be mathematicians if they did otherwise!)
Plenty of free online resources will calculate the line of best ft for you, if you provide your set of data. I like wolfram alpha, it can do almost anythng! Check it out when you have a spare hour (or six): http://www.wolframalpha.com
this site might be useful too :)
http://www.alcula.com/calculators/statistics/linear-regression/