One imagines the joint normal distribution for two variables as a 3D bell. Looking from the top, one makes a horizontal cut at the 95% contour. The resulting shape is an elipse. If the variables are independent and standardized, the contour will be a circle. One tests to see if a sample point x1,y1 plots outside the elipse.

There are several strategies to compute confidence intervals (asymptotic, Monte-Carlo based method, bootstrap and F-test based confidence assessment technique). Please refer this book: http://www.graphpad.com/manuals/prism4/RegressionBook.pdf

It will give details about all the techniques. If you are using Matlab: http://de.mathworks.com/help/curvefit/confidence-and-prediction-bounds.html

I hope this link may help you. (R regression

http://www.cyclismo.org/tutorial/R/confidence.html