Thank you but sorry. this is not what I want, please see PDF file I attached previously.

All right, finally I could see your example, without actual plot. If you are so picky on the details you have to do it by hand. It is possible that there is a package hiding somewhere which includes a secret function that can do it. But I personally find it much quicker to write some scripts as following:

And output:



There a 1-character width space between the text and the left margin of the plot and between the text and the symbol. The line spacing is 1.5 times character height. The space between top margin of the plot and the legend is half of a character height. And the distance between points in the legend is 1-character width.

Anyway I agree with qioqpwqr, single point in the legend seems enough. Hard to imagine cases that you need definitely this type of legend.

First of all, Thank you very much! very big progress! but still need one more help.

Actually, for the lengend, what I need is only one lengend with three different colors of points instead of three legends, I mean, for example, for y1 in the legend in your example, I need colors of points are green, red and blue on one line, instead of three legends with different colors and color of points are all same for each legend. There is only one line connecting three groups of points with three different colors of green, red, blue in the plot. But i don't need generate plot, I already finished generating plot. I just need legend. In the example in my PDF file, I need change three black points to green, red, blue for that one legend only, no second and third legend please.  

I think this is hard, and don't know how to generate this legend with R. but I am a little bit excited we are closing.

Finally you explain your wish CLEARLY. If you accept using low level functions to customize your legend you could do pretty much what you want. It should not be hard if you understand the example I have shown above.
Output:




I completely ignore the plot as I was not able to imagine a plot that needs such legend... Will you show your plot, perhaps with dummy data?
The way to determine the location of the symbols has been changed. It is more precise to use the strwidth function than nchar()*par()$cxy[1] according to R's documentation.

Wonderful help! I appreciate! This is exactly what I want! Thank you again!